Sound FontÒ Technical Specification Version 2.00a October 18, 1995 0 About This Document 0.1 Revision History Revision Issue Date Comments 2.00a 10/18/95 First publicly released draft 0.2 Disclaimers THIS SPECIFICATION IS PROVIDED ÒAS ISÓ WITH NO WARRANTIES WHATSOEVER INCLUDING ANY WARRANTY OF MERCHANTABILITY, FITNESS FOR ANY PARTICULAR PURPOSE, OR ANY WARRANTEE OTHERWISE ARISING OUT OF ANY PROPOSAL, SPECIFICATION, OR SAMPLE. A LICENSE IS HEREBY GRANTED TO COPY, REPRODUCE, AND DISTRIBUTE THIS SPECIFICATION FOR INTERNAL USE ONLY. NO OTHER LICENSE EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY OTHER INTELLECTUAL PROPERTY RIGHTS IS GRANTED OR INTENDED HEREBY. AUTHORS OF THIS SPECIFICATION DISCLAIM ALL LIABILITY, INCLUDING LIABILITY FOR INFRINGEMENT OF PROPPIETARY RIGHTS, RELATING TO IMPLEMENTATION OF INFORMATION IN THIS SPECIFICATION. AUTHORS OF THIS SPECIFICATION ALSO DO NOT WARRANT OR REPRESENT THAT SUCH IMPLEMENTATION(S) WILL NOT INFRINGE ON SUCH RIGHTS. This preliminary document is being distributed solely for the purpose of review and solicitation of comments. It will be updated periodically. No products should rely on the content of this version of the document. SoundFontÒ is a registered trademark of E-mu Systems, Inc. E-mu Systems licenses a ÒSoundFont CompatibilityÓ logo for a nominal fee; please contact E-muÕs SoundFont administrator by FAX at (408) 439-0392 for more information. Users of the information contained herein should refer to files conforming to the specification as ÒSoundFont Compatible,Ó with appropriate acknowledgement of trademark ownership. 0.3 Comments Please send comments via e-mail to soundfont@emu.com 0.4 Table of Contents 0 About This Document 0.1 Revision History 0.2 Disclaimers 0.3 Comments 0.4 Table of Contents 0.5 Illustrations 1 Introduction 1.1 Scope and Intended Purpose of this Document 1.2 Document Organization 1.3 SoundFont 2 Objectives 1.4 SoundFont 1.x 1.5 Future Enhancements to the SoundFont 2 Standard 2 Terms and Abbreviations 2.1 Data Structure Terminology 2.2 Synthesizer Terminology 2.3 Parameter Terminology 3 RIFF Structure 3.1 General RIFF File Structure 3.2 The SoundFont 2 Chunks and Subchunks 3.3 Redundancy and Error Handling in the RIFF structure 4 SoundFont 2 RIFF File Format 4.1 SoundFont 2 RIFF File Format Level 0 4.2 SoundFont 2 RIFF File Format Level 1 4.3 SoundFont 2 RIFF File Format Level 2 4.4 SoundFont 2 RIFF File Format Level 3 4.5 SoundFont 2 RIFF File Format Type Definitions 5 The INFO-list Chunk 5.1 The ifil Subchunk 5.2 The isng Subchunk 5.3 The INAM Subchunk 5.4 The irom Subchunk 5.5 The iver Subchunk 5.6 The ICRD Subchunk 5.7 The IENG Subchunk 5.8 The IPRD Subchunk 5.9 The ICOP Subchunk 5.10 The ICMT Subchunk 5.11 The ISFT Subchunk 6 The sdta-list Chunk 6.1 Sample Data Format in the smpl Subchunk 6.2 Sample Data Looping Rules 7 The pdta-list Chunk 7.1 The HYDRA Data Structure 7.2 The PHDR Subchunk 7.3 The PBAG Subchunk 7.4 The PMOD Subchunk 7.5 The PGEN Subchunk 7.6 The INST Subchunk 7.7 The IBAG Subchunk 7.8 The IMOD Subchunk 7.9 The IGEN Subchunk 7.10 The SHDR Subchunk 8 Enumerators 8.1 Generator Enumerators 8.1.1 Kinds of Generator Enumerators 8.1.2 Generator Enumerators Defined 8.1.3 Generator Summary 8.2 Source Enumerators 8.3 Transform Enumerators 8.4 Default Modulators 8.5 Precedence and Absolute and Relative values. 9 Parameters and Synthesis Model 9.1 Synthesis Model 9.1.1 Wavetable Oscillator 9.1.2 Sample Looping 9.1.3 Lowpass Filter 9.1.4 Final Gain Amplifier 9.1.5 Effects Sends 9.1.6 Low Frequency Oscillators 9.1.7 Envelope Generators 9.1.8 Modulation Interconnection Summary 9.2 MIDI Functions 9.3 Parameter Units 9.4 On Implementation Accuracy 10 Error Handling 10.1 Structural Errors 10.2 Unknown Chunks 10.3 Unknown Enumerators 10.4 Illegal Parameter Values 10.5 Unusual Values 10.6 Missing Required Parameter or Terminator 10.7 Illegal enumerator 11 Silicon SoundFonts 12 Glossary 0.5 Illustrations Figure 1 - Ideal Filter Response Section 9.1.3 Figure 2 - Modulation Structure Section 9.1.8 1 Introduction 1.1 Scope and Intended Purpose of this Document This document is the definitive source for the SoundFont 2 standard. This document should provide complete and accurate information to allow any user to correctly construct and interpret SoundFont 2 compatible banks. This document is not intended to provide any information on the design or implementation of music synthesizers. 1.2 Document Organization This document is organized such that sections 1 and 2 give introductory information about the SoundFont 2 standard. Both new and seasoned musical engineers will get value from the review of terminology provided in section 2. Sections 3 through 8 provide increasingly detailed descriptions of the SoundFont 2 standard data structures. The sections will ultimately serve as reference, but can be scanned in order to provide sufficient detail for any level of understanding. Section 9 deals with the Synthesis model supported by the SoundFont standard, and will be of interest to anyone involved with the synthesis engine or bank creation. Section 10 specifies error handling when dealing with SoundFont compatible banks, and will be of interest primarily to programmers using the SoundFont standard. The alphabetical glossary in section 11 can be used as a reference for any unfamiliar or confusing terminology. 1.3 SoundFont 2 Objectives The SoundFont 2 standard is intended to provide an extensible, portable, universal interchange format for wavetable synthesizer ÒsamplesÓ and articulation data. The standard is made extensible largely by the use of enumerated ÒgeneratorsÓ and ÒmodulatorsÓ so that additional function units can be added as requirements dictate. The standard is made portable and universal by the use of precisely defined and hardware independent parameters, as well as by specific practices designed to provide support to a broad range of technologies. 1.4 SoundFont 1.x The SoundFont standard was originally released in its 1.0 embodiment with the Creative Technology AWE32 product using the EMU8000 music synthesis chip. This proprietary format proved very successful, but experience brought a number of refinements. These initially were performed in an upward compatible manner to revision 1.5. However, due to increasing demand for a public downloadable sound interchange format, Creative Technology determined that a public disclosure of the SoundFont format would be in its best interest. Because there were still more improvements required, many of which could not be supported in a completely compatible manner, Creative decided to combine public disclosure with the step to a revised format. The result is the SoundFont 2 standard. There are several key enhancements contained in the SoundFont 2 standard. The first is the use of relative parameters in the Preset level. This allows instruments to be adjusted without altering their self-consistency, providing easy and effective user editing of instruments. The second is an improvement in the data structures associated with the samples themselves, again providing key information which will allow the sound designer to re-use samples with a minimum of difficulty. An increased specificity in the rules for sample data produces enhanced portability across various sound engines. Finally, the addition of modulators produces a robust structure which can express all the typical function in current and future wavetable synthesizers. 1.5 Future Enhancements to the SoundFont 2 Standard The SoundFont 2 standard is designed to allow for enhancements based on future wavetable synthesis technology capabilities by additional enumerations of generators and modulators. This will be done as required in an upwardly compatible manner. Suggestions for additions can be made via e-mail to soundfont@emu.com. In general, our policy for updating the specification will be based on consumer need, rather than technological idealism. It is our expectation to maintain bidirectional compatibility within the SoundFont 2 standard for some years. 2 Terms and Abbreviations The following sections introduce terms used within this specification in a logical order. They are provided both as an introduction to readers unfamiliar with wavetable synthesis implementation details, as well as a review and reference for the expert. These and other terms and abbreviations can also be found arranged alphabetically for reference in the glossary at the end of this specification. 2.1 Data Structure Terminology bag - A SoundFont data structure element containing a list of layers (preset bag) or splits (instrument bag). big endian - Refers to the organization in memory of bytes within a word such that the most significant byte occurs at the lowest address. Contrast Òlittle endian.Ó byte - A data structure element of eight bits without definition of meaning to those bits. BYTE - A data structure element of eight bits which contains an unsigned value from 0 to 255. case-insensitive - Indicates that an ASCII character or string treats alphabetic characters of upper or lower case as identical. Contrast Òcase-sensitive.Ó case-sensitive - Indicates that an ASCII character or string treats alphabetic characters of upper or lower case as distinct. Contrast Òcase-insensitive.Ó CHAR - A data structure of eight bits which contains a signed value from -128 to +127. chunk - The top level division of a RIFF file. doubleword - A data structure element of 32 bits without definition of meaning to those bits. DWORD - A data structure of 32 bits which contains an unsigned value from zero to 4,294,967,295. enumerated - Said of a data element whose symbols correspond to particular assigned functions. global - Refers to parameters which affect all associated structures. See Òglobal layerÓ and Òglobal split.Ó global layer - A layer whose generators and modulators affect all other layers within the preset. global split - A split whose generators and modulators affect all other splits within the instrument. header - A data structure element which describes several aspects of a SoundFont element. hydra - A. A nine-headed mythical beast. B. The nine ÒpdtaÓ subchunks which make up the SoundFont articulation data. instrument - In the SoundFont standard, a collection of splits which represents the sound of a single musical instrument or sound effect set. instrument split - A sample and associated articulation data defined to play over certain key numbers and velocities. Also simply called a split. layer - A subset of a preset containing generators, modulators, and an instrument. Also termed Òpreset layer.Ó level - In the SoundFont structure, this refers either to the preset and layers (the preset level) or the instrument and splits (the instrument level). little endian - A method of ordering bytes within larger words in memory in which the least significant byte is at the lowest address. Contrast Òbig endian.Ó orphan - Said of a data structure which under normal circumstances is referenced by a higher level, but in this particular instance is no longer linked. Specifically, it is an instrument which is not referenced by any preset layer, or a sample which is not referenced by any instrument split. preset - A keyboard full of sound. Typically the collection of samples and articulation data associated with a particular MIDI preset number. preset layer - A subset of a preset containing generators, modulators, and an instrument. Also simply termed a Òlayer.Ó record - A single instrance of a data structure. RIFF - Acronym for Resource Interchange File Format. The recommended form for interchange files such as SoundFont compatible files within Microsoft operating systems. SHORT - A data structure element of sixteen bits which contains a signed value from -32,768 to +32,767. split - A sample and associated articulation data defined to play over certain key numbers and velocities. Also called an instrument split. subchunk - A division of a RIFF file below that of the chunk. terminator - A data structure element indicating the final element in a sequence. WORD - A data structure of 16 bits which contains an unsigned value from zero to 65,535. word - A data structure element of 16 bits without definition of meaning to those bits. 2.2 Synthesizer Terminology articulation - The process of modulation of amplitude, pitch, and timbre to produce an expressive musical note. artifact - A (typically undesirable) sonic event which is recognizable as not being present in the original sound. attack - That phase of an envelope or sound during which the amplitude increases from zero to a peak value. attenuation - A decrease in volume or amplitude of a signal. AWE32 - The original Creative Technology Sound Blaster product which contained an EMU8000 wavetable synthesizer and supported the SoundFont standard. balance - A form of stereo volume control in which both left and right channels are at maximum when the control is centered, and which attenuates only the opposite channel when taken to either extreme. bank - A collection of presets. See also MIDI bank. chorus - An effects processing algorithm which involves cyclically shifting the pitch of a signal and remixing it with itself to produce a time varying comb filter, giving a perception of motion and fullness to the resulting sound. cutoff frequency - The frequency of a filter function at which the attenuation reaches a specified value. data points - The individual values comprising a sample. Sometimes also called sample points. Contrast Òsample.Ó decay - The portion of an envelope or sound during which the amplitude declines from a peak to steady state value. delay - The portion of an envelope or LFO function which elapses from a key-on event until the amplitude becomes non-zero. DC gain - The degree of amplification or attentuation a system presents to a static or zero frequency signal. digital audio - Audio represented as a sequence of quantized values spaced evenly over time. The values are called Òsample data points.Ó downloadable - Said of samples which are loaded from a file into RAM, in contrast to samples which are maintained in ROM. dry - Refers to audio which has not received any effects processing such as reverb or chorus. EMU8000 - A wavetable synthesizer chip designed by E-mu Systems for use in Creative Technology products. envelope - A time varying signal which typically controls the pitch, volume, and/or filter cutoff frequency of a note, and comprises multiple phases including attack, decay, sustain, and release. flat - A. Said of a tone that is lower in pitch than another reference tone. B. Said of a frequency response that does not deviate significantly from a single fixed gain over the audio range. interpolator - A circuit or algorithm which computes intermediate points between existing sample data points. This is of particular use in the pitch shifting operation of a wavetable synthesizer, in which these intermediate points represent the output samples of the waveform at the desired pitch transposition. key number - See MIDI key number. LFO - Acronym for Low Frequency Oscillator. A slow periodic modulation source. linear coding - The most common method of encoding amplitudes in digital audio in which each step is of equal size. loop - In wavetable synthesis, a portion of a sample which is repeated many times to increase the duration of the resulting sound. loop points - The sample data points at which a loop begins and ends. lowpass - Said of a filter which attenuates high frequencies but does not attenuate low frequencies. MIDI - Acronym for Musical Instrument Digital Interface. The standard protocol for sending performance information to a musical synthesizer. MIDI bank - A group of up to 128 presets selected by a MIDI Òchange bankÓ command. MIDI continuous controller - A construct in the MIDI protocol. MIDI key number - A construct in the MIDI protocol which accompanies a MIDI key-on or key-off command and specifies the key of the musical instrument keyboard to which the command refers. MIDI pitch bend - A special MIDI construct akin to the MIDI continuous controllers which controls the realtime value of the pitch of all notes played in a MIDI channel. MIDI preset - A ÒpresetÓ selected to be active in a particular MIDI channel by a MIDI Òchange presetÓ command. MIDI velocity - A construct in the MIDI protocol which accompanies a MIDI key-on or key-off command and specifies the speed with which the key was pressed or released. mono - Short for Òmonophonic.Ó Indicates a sound comprising only one channel or waveform. Contrast with Òstereo.Ó oscillator - In wavetable synthesis, the wavetable interpolator is considered an oscillator. pan - Short for Òpanorama.Ó This is the control of the apparent azimuth of a sound source over 180 degrees from left to right. It is generally implemented by varying the volume at the left and right speakers. pitch - The perceived value of frequency. Generally can be used interchangably with frequency. pitch shift - A change in pitch. Wavetable synthesis relies on interpolators to cause pitch shift in a sample to produce the notes of the scale. pole - A mathematical term used in filter transform analysis. Traditionally in synthesis, a pole is equated with a rolloff of 6dB per octave, and the rolloff of a filter is specified in Òpoles.Ó Preditor - E-mu SystemsÕ proprietary SoundFont 2.00 compatible bank editing software. preset - A keyboard full of sound. Typically the collection of samples and articulation data associated with a particular MIDI preset number. Q - A mathematical term used in filter transform analysis. Indicates the degree of resonance of the filter. In synthesis terminology, it is synonymous with resonance. release - The portion of an envelope or sound during which the amplitude declines from a steady state to zero value or inaudibility. resonance - Describes the aspect of a filter in which particular frequencies are given significantly more gain than others. The resonance can be measured in dB above the DC gain. resonant frequency - The frequency at which resonance reaches its maximum. reverb - Short for reverberation. In synthesis, a synthetic signal processor which adds artificial spaciousness and ambience to a sound. sample - This term is often used both to indicate a Òsample data pointÓ and to indicate a collection of such points comprising a digital audio waveform. The latter meaning is exclusively used in this specification. soft - The pedal on a piano, so named because it causes the damper to be lowered in such a way as to soften the timbre and loudness of the notes. In MIDI, continuous controller #66 which behaves in a similar manner. sostenudo - The pedal on a piano which causes the dampers on all keys depressed to be held until the pedal is released. In MIDI, continuous controller #67 which behaves in a similar manner. sustain - The pedal on a piano which prevents all dampers on keys as they are depressed from being released. In MIDI, continuous controller #64 which behaves in a similar manner. SoundFont - A registered trademark of E-mu Systems, Inc, indicating files produced by E-mu which conform to the SoundFont standard file format. stereo - Literally indicating three dimensions. In this specification, the term is used to mean two channel stereophonic, indicating that the sound is composed of two independent audio channels, dubbed left and right. Constrast monophonic. synthesis engine - The hardware and software associated with the signal processing and modulation path for a particular synthesizer. synthesizer - A device capable of producing ideally arbitrary musical sound. tremolo - A periodic change in amplitude of a sound, typically produced by applying a low frequency oscillator to the final volume amplifier. triangular - A waveform which ramps upward to a positive limit, then downward at the opposite slope to the symmetrically negative limit periodically. unpitched - Said of a sound which is not characterized by a perceived frequency. This would be true of noise-like musical instruments and of many sound effects. velocity - In synthesis, the speed with which a keyboard key is depressed, typically proportionally to the impact delivered by the musician. See also MIDI velocity. vibrato - A periodic change in the pitch of a sound, typically produced by applying a low frequency oscillator to the oscillator pitch. volume - The loudness or amplitude of a sound, or the control of this parameter. wavetable - A music synthesis technique wherein musical sounds are recorded or computed mathematically and stored in a memory, then played back at a variable rate to produce the desired pitch. Additional timbral adjustments are often made to the sound thus produced using amplifiers, filters, and effect processing such as reverb and chorus. 2.3 Parameter Terminology absolute - Describes a parameter which gives a definitive real-world value. Contrast to relative. additive - Describes a parameter which is to be numerically added to another parameter. attenuation - A decrease in volume or amplitude of a signal. cent - A unit of pitch ratio corresponding to the twelve hundredth root of two, or one hundredth of a semitone, approximately 1.000577790. centibel - A unit of amplitude ratio corresponding to the two hundredth root of ten, or one tenth of a decibel, approximately 1.011579454. cutoff frequency - The frequency of a filter function at which the attenuation reaches a specified value. decibel - A unit of amplitude ratio corresponding to the twentieth root of ten, approximately 1.122018454. octave - A factor of two in ratio, typically applied to pitch or frequency. pitch - The perceived value of frequency. Generally can be used interchangably with frequency. pitch shift - A change in pitch. Wavetable synthesis relies on interpolators to cause pitch shift in a sample to produce the notes of the scale. relative - Describes a parameter which merely indicates an offset from an otherwise established value. Contrast to absolute. resonance - Describes the aspect of a filter in which particular frequencies are given significantly more gain than others. The resonance can be measured in dB above the DC gain. sample rate - The frequency, in Hertz, at which sample data points are taken when recording a sample. semitone - A unit of pitch ratio corresponding to the twelfth root of two, or one twelfth of an octave, approximately 1.059463094. sharp - Said of a tone that is higher in pitch than another reference tone. timecent - A unit of duration ratio corresponding to the twelve hundredth root of two, or one twelve hundredth of an octave, approximately 1.000577790. 3 RIFF Structure 3.1 General RIFF File Structure The RIFF (Resource Interchange File Format) is a tagged file structure developed for multimedia resource files, and is described in some detail in the Microsoft Windows 3.1 SDK Multimedia ProgrammerÕs Reference. the Tagged-file structure is useful because it helps prevent compatibility problems which can occur as the file definition changes over time. Because each piece of data in the file is identified by a standard header, an application that does not recognize a given data element can skip over the unknown information. A RIFF file is constructed from a basic building block called a Òchunk.Ó In ÔCÕ syntax, a chunk is defined: typedef DWORD FOURCC; // Four-character code typedef struct { FOURCC ckID; // A chunk ID identifies the type of data within the chunk. DWORD ckSize; // The size of the chunk data in bytes, excluding any pad byte. BYTE ckDATA[ckSize]; // The actual data plus a pad byte if reqÕd to word align. }; Two types of chunks, the ÒRIFFÓ and ÒLISTÓ chunks, may contain nested chunks called subchunks as their data. The ordering requirements of chunks and subchunks within a RIFF file is not well documented in the RIFF file format. In SoundFont 2.0, the order of the subchunks withing the INFO chunk is arbitrary, but for consistency it is recommended that the subchunks be ordered as presented in this document. The order of the all other chunks and subchunks is strictly defined and must be maintained as presented in this document. 3.2 The SoundFont 2 Chunks and Subchunks A SoundFont 2 compatible RIFF file comprises three chunks: an INFO-list chunk containing a number of required and optional subchunks describing the file, its history, and its intended use, an sdta-list chunk comprising a single subchunk containing any referenced digital audio samples, and a pdta-list chunk containing nine subchunks which define the articulation of the digital audio data. The SoundFont 2 standard allows that the subchunks within the INFO-list chunk may appear in arbitrary order. However, the order of the three chunks, and the order of the subchunks within the pdta- list chunk, is fixed. The SoundFont 2 specification requires that implementations ignore unknown subchunks within the INFO-list chunk. Note, however, that until such subchunks become defined in the specification, inclusion of additional INFO-list subchunks will preclude the file from conforming to the SoundFont standard. A detailed description of the SoundFont 2 RIFF structure is provided in Section 4. 3.3 Redundancy and Error Handling in the RIFF structure The RIFF file structure contains redundant information regarding the length of the file and the length of the chunks and subchunks. This fact enables any reader of a SoundFont compatible file to determine if the file has been damaged by loss of data. If any such loss is detected, the SoundFont compatible file is termed Òstructurally unsoundÓ and in general should be rejected. SoundFont compatible software developers may produce utilities to recover data from structurally unsound files, producing with or without user assitance a corrected and structurally sound SoundFont 2 compatible file. 4 SoundFont 2 RIFF File Format 4.1 SoundFont 2 RIFF File Format Level 0 -> RIFF (ÔsfbkÕ ; RIFF form header { ; Supplemental Information ; The Sample Binary Data ; The Preset, Instrument, and Sample Header data } ) 4.2 SoundFont 2 RIFF File Format Level 1 -> LIST (ÔINFOÕ { ; Refers to the version of the Sound Font RIFF file ; Refers to the target Sound Engine ; Refers to the Sound Font Bank Name [] ; Refers to the Sound ROM Name [] ; Refers to the Sound ROM Version [] ; Refers to the Date of Creation of the Bank [] ; Sound Designers and Engineers for the Bank [] ; Product for which the Bank was intended [] ; Contains any Copyright message [] ; Contains any Comments on the Bank [] ; The SoundFont tools used to create and alter the bank } ) -> LIST (ÔsdtaÕ { [ -> LIST (ÔpdtaÕ { ; The Preset Headers ; The Preset Index list ; The Preset Modulator list ; The Preset Generator list ; The Instrument Names and Indicies ; The Instrument Index list ; The Instrument Modulator list ; The Instrument Generator list ; The Sample Headers } ) 4.3 SoundFont 2 RIFF File Format Level 2 -> ifil() ; e.g. 2.00 -> isng(szSoundEngine:ZSTR) ; e.g. ÒEMU8000Ó -> irom(szROM:ZSTR) ; e.g. Ò1MGMÓ -> iver() ; e.g. 2.08 -> INAM(szName:ZSTR) ; e.g. ÒGeneral MIDIÓ -> ICRD(szDate:ZSTR) ; e.g. ÒJuly 15, 1995Ó -> IENG(szName:ZSTR) ; e.g. ÒJohn Q. EngineerÓ -> IPRD(szProduct:ZSTR) ; e.g. ÒSBAWE32Ó -> ICOP(szCopyright:ZSTR) ; e.g. ÒCopyright (c) 1995 E-mu Systems, Inc.Ó -> ICMT(szComment:ZSTR) ; e.g. ÒThis is a commentÓ -> ISFT(szTools:ZSTR) ; e.g. ÒPreditor 2.00a:Preditor 2.00aÓ -> smpl() ; 16 bit Linearly Coded Digital Audio Data -> phdr() -> pbag() -> pmod() -> pgen() -> inst () -> ibag() -> imod() -> igen() -> shdr() 4.4 SoundFont 2 RIFF File Format Level 3 -> struct sfVersionTag { WORD wMajor; WORD wMinor; }; -> struct sfPresetHeader { CHAR achPresetName[20]; WORD wPreset; WORD wBank; WORD wPresetBagNdx; DWORD dwLibrary; DWORD dwGenre; DWORD dwMorphology; }; -> struct sfPresetBag { WORD wGenNdx; WORD wModNdx; }; -> struct sfModList { SFModulator sfModSrcOper; SFGenerator sfModDestOper; SHORT modAmount; SFModulator sfModAmtSrcOper; SFTransform sfModTransOper; }; -> struct sfGenList { SFGenerator sfGenOper; genAmountType genAmount; }; -> struct sfInst { CHAR achInstName[20]; WORD wInstBagNdx; }; -> struct sfInstBag { WORD wInstGenNdx; WORD wInstModNdx; }; -> struct sfInstModList { SFModulator sfModSrcOper; SFGenerator sfModDestOper; SHORT modAmount; SFModulator sfModAmtSrcOper; SFTransform sfModTransOper; }; -> struct sfInstGenList { SFGenerator sfGenOper; genAmountType genAmount; }; -> struct sfSample { CHAR achSampleName[20]; DWORD dwStart; DWORD dwEnd; DWORD dwStartloop; DWORD dwEndloop; DWORD dwSampleRate; BYTE byOriginalKey; CHAR chCorrection; WORD wSampleLink; SFSampleLink sfSampleType; }; 4.5 SoundFont 2 RIFF File Format Type Definitions The sfModulator, sfGenerator, and sfTransform types are all enumeration types whose values are defined in subsequent sections. The genAmountType is a union which allows signed 16 bit, unsigned 16 bit, and two unsigned 8 bit fields: typedef struct { BYTE byLo; BYTE byHi; } rangesType; typedef union { rangesType ranges; SHORT shAmount; WORD wAmount; } genAmountType; The SFSampleLink is an enumeration type which describes both the type of sample (mono, stereo left, etc.) and the whether the sample is located in RAM or ROM memory: typedef enum { monoSample = 1, rightSample = 2, leftSample = 4, linkedSample = 8, RomMonoSample = 0x8001, RomRightSample = 0x8002, RomLeftSample = 0x8004, RomLinkedSample = 0x8008 } SFSampleLink; 5 The INFO-list Chunk TheINFO-list chunk in a SoundFont 2 compatible file contains three mandatory and a variety of optional subchunks as defined below. The INFO-list chunk gives basic information about the SoundFont compatible bank contained in the file. 5.1 The ifil Subchunk The ifil subchunk is a mandatory subchunk identifying the SoundFont specification version level to which the file complies. It is always four bytes in length, and contains data according to the structure: struct sfVersionTag { WORD wMajor; WORD wMinor; }; The WORD wMajor contains the value to the left of the decimal point in the SoundFont specification version, the WORD wMinor contains the value to the right of the decimal point. For example, version 2.11 would be implied if wMajor=2 and wMinor=11. These values can be used by applications which read SoundFont compatible files to determine if the format of the file is usable by the program. Within a fixed wMajor, the only changes to the format will be the addition of Generator, Source and Transform enumerators, and additional info subchunks. These are all defined as being ignored if unknown to the program. Consequently, many applications can be designed to be fully upward compatible within a given wMajor. In the case of editors or other programs in which all enumerators should be known, the value of wMinor may be of consequence. Generally the application program will either accept the file as usable (possibly with appropriate transparent translation), reject the file as unusable, or warn the user that there may be uneditable data in the file. If the ifil subchunk is missing, or its size is not four bytes, the file should be rejected as structurally unsound. 5.2 The isng Subchunk The isng subchunk is a mandatory subchunk identifying the wavetable sound engine for which the file was optimized. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. The default isng field is the eight bytes representing ÒEMU8000Ó as seven ASCII characters followed by a zero byte. The ASCII should be treated as case-sensitive. In other words Òemu8000Ó is not the same as ÒEMU8000.Ó The isng string can be optionally used by chip drivers to vary their synthesis algorithms to emulate the target sound engine. If the isng subchunk is missing, not terminated in a zero valued byte, or its contents are an unknown sound engine, the field should be ignored and EMU8000 assumed. 5.3 The INAM Subchunk The INAM subchunk is a mandatory subchunk providing the name of the SoundFont compatible bank. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical inam subchunk would be the fourteen bytes representing ÒGeneral MIDIÓ as twelve ASCII characters followed by two zero bytes. The ASCII should be treated as case-sensitive. In other words ÒGeneral MIDIÓ is not the same as ÒGENERAL MIDI.Ó The inam string is typically used for the idenitification of banks even if the file names are altered. If the inam subchunk is missing, or not terminated in a zero valued byte, the field should be ignored and the user supplied with an appropriate error message if the name is queried. If the file is re-written, a valid name should be placed in the INAM field. 5.4 The irom Subchunk The irom subchunk is an optional subchunk identifying a particular wavetable sound data ROM to which any ROM samples refer. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical irom field would be the six bytes representing Ò1MGMÓ as four ASCII characters followed by two zero bytes. The ASCII should be treated as case-sensitive. In other words Ò1mgmÓ is not the same as Ò1MGM.Ó The irom string is used by drivers to verify that the ROM data referenced by the file is available to the sound engine. If the irom subchunk is missing, not terminated in a zero valued byte, or its contents are an unknown ROM, the field should be ignored and the file assumed to reference no ROM samples. If ROM samples are accessed, any accesses to such intruments should be terminated and not sound. A file should not be written which attempts to access ROM samples without both irom and iver present and valid. 5.5 The iver Subchunk The iver subchunk is an optional subchunk identifying the particular wavetable sound data ROM revision to which any ROM samples refer. It is always four bytes in length, and contains data according to the structure: struct sfVersionTag { WORD wMajor; WORD wMinor; }; The WORD wMajor contains the value to the left of the decimal point in the ROM version, the WORD wMinor contains the value to the right of the decimal point. For example, version 1.36 would be implied if wMajor=1 and wMinor=36. The iver subchunk is used by drivers to verify that the ROM data referenced by the file is located in the exact locations specified by the sound headers. If the iver subchunk is missing, not four bytes in length, or its contents indicate an unknown or incorrect ROM, the field should be ignored and the file assumed to reference no ROM samples. If ROM samples are accessed, any accesses to such instruments should be terminated and not sound. Note that for ROM samples to function correctly, both iver and irom must be present and valid. A file should not be written which attempts to access ROM samples without both irom and iver present and valid. 5.6 The ICRD Subchunk The ICRD subchunk is an optional subchunk identifying the creation date of the SoundFont compatible bank. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical ICRD field would be the twelve bytes representing ÒMay 1, 1995Ó as eleven ASCII characters followed by a zero byte. Conventionally, the format of the string is ÒMonth Day, YearÓ where Month is initially capitalized and is the conventional full English spelling of the month, Day is the date in decimal followed by a comma, and Year is the full decimal year. Thus the field should conventionally never be longer than 32 bytes. The ICRD string is provided for library management purposes. If the ICRD subchunk is missing, not terminated in a zero valued byte, or for some reason incapable of being faithfully copied as an ASCII string, the field should be ignored and if re-written, should not be copied. If the fieldÕs contents are not seemingly meaningful but can faithfully reproduced, this should be done. 5.7 The IENG Subchunk The IENG subchunk is an optional subchunk identifying the names of any sound designers or engineers responsible for the SoundFont compatible bank. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical IENG field would be the twelve bytes representing ÒTim SwartzÓ as ten ASCII characters followed by two zero bytes. The IENG string is provided for library management purposes. If the IENG subchunk is missing, not terminated in a zero valued byte, or for some reason incapable of being faithfully copied as an ASCII string, the field should be ignored and if re-written, should not be copied. If the fieldÕs contents are not seemingly meaningful but can faithfully reproduced, this should be done. 5.8 The IPRD Subchunk The IPRD subchunk is an optional subchunk identifying any specific product for which the SoundFont compatible bank is intended. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical IPRD field would be the eight bytes representing ÒSBAWE32Ó as seven ASCII characters followed by a zero byte. The ASCII should be treated as case-sensitive. In other words Òsbawe32Ó is not the same as ÒSBAWE32.Ó The IPRD string is provided for library management purposes. If the IPRD subchunk is missing, not terminated in a zero valued byte, or for some reason incapable of being faithfully copied as an ASCII string, the field should be ignored and if re-written, should not be copied. If the fieldÕs contents are not seemingly meaningful but can faithfully reproduced, this should be done. 5.9 The ICOP Subchunk The ICOP subchunk is an optional subchunk containing any copyright assertion string associated with the SoundFont compatible bank. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical ICOP field would be the 40 bytes representing ÒCopyright (c) 1995 E-mu Systems, Inc.Ó as 38 ASCII characters followed by two zero bytes. The ICOP string is provided for intellectual property protection and management purposes. If the ICOP subchunk is missing, not terminated in a zero valued byte, or for some reason incapable of being faithfully copied as an ASCII string, the field should be ignored and if re-written, should not be copied. If the fieldÕs contents are not seemingly meaningful but can faithfully reproduced, this should be done. 5.10 The ICMT Subchunk The ICMT subchunk is an optional subchunk containing any comments associated with the SoundFont compatible bank. It contains an ASCII string of 65,536 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical ICMT field would be the 40 bytes representing ÒThis space unintentionally left blank.Ó as 38 ASCII characters followed by two zero bytes. The ICMT string is provided for any non-scatological uses. If the ICMT subchunk is missing, not terminated in a zero valued byte, or for some reason incapable of being faithfully copied as an ASCII string, the field should be ignored and if re-written, should not be copied. If the fieldÕs contents are not seemingly meaningful but can faithfully reproduced, this should be done. 5.11 The ISFT Subchunk The ISFT subchunk is an optional subchunk identifying the SoundFont compatible tools used to create and most recently modify the SoundFont compatible bank. It contains an ASCII string of 256 or fewer bytes including one or two terminators of value zero, so as to make the total byte count even. A typical ISFT field would be the thirty bytes representing ÒPreditor 2.00a:Preditor 2.00aÓ as twenty-nine ASCII characters followed by a zero byte. The ASCII should be treated as case-sensitive. In other words ÒPreditorÓ is not the same as ÒPREDITOR.Ó Conventionally, the tool name and revision control number are included first for the creating tool and then for the most recent modifying tool. The two strings are separated by a colon. The string should be produced by the creating program with a null modifying tool field (e.g. ÒPreditor 2.00a:), and each time a tool modifies the bank, it should replace the modifying tool field with its own name and revision control number. The ISFT string is provided primarily for error tracing purposes. If the ISFT subchunk is missing, not terminated in a zero valued byte, or for some reason incapable of being faithfully copied as an ASCII string, the field should be ignored and if re-written, should not be copied. If the fieldÕs contents are not seemingly meaningful but can faithfully reproduced, this should be done. 6 The sdta-list Chunk The sdta-list chunk in a SoundFont 2 compatible file contains a single optional smpl subchunk which contains all the RAM based sound data associated with the SoundFont compatible bank. The smpl subchunk is of arbitrary length, and contains an even number of bytes. 6.1 Sample Data Format in the smpl Subchunk The smpl subchunk, if present, contains one or more ÒsamplesÓ of digital audio information in the form of linearly coded sixteen bit, signed, little endian (least significant byte first) words. Each sample is followed by a minimum of forty-six zero valued sample data points. These zero valued data points are necessary to guarantee that any reasonable upward pitch shift using any reasonable interpolator can loop on zero data at the end of the sound. 6.2 Sample Data Looping Rules Within each sample, one or more loop point pairs may exist. The locations of these points are defined within the pdta-list chunk, but the sample data points themselves must comply with certain practices in order for the loop to be compatible across multiple platforms. The loops are defined by Òequivalent pointsÓ in the sample. This means that there are two sample data points which are logically equivalent, and a loop occurs when these points are spliced atop one another. In concept, the loop end point is never actually played during looping; instead the loop start point follows the point just prior to the loop end point. Because of the bandlimited nature of digital audio sampling, an artifact free loop will exhibit virtually identical data surrounding the equivalent points. In actuality, because of the various interpolation algorithms used by wavetable synthesizers, the data surrounding both the loop start and end points may affect the sound of the loop. Hence both the loop start and end points must be surrounded by continuous audio data. For example, even if the sound is programmed to continue to loop throughout the decay, sample data points must be provided beyond the loop end point. This data will typically be identical to the data at the start of the loop. A minimum of eight valid data points are required to be present before the loop start and after the loop end. The eight data points (four on each side) surrounding the two equivalent loop points should also be forced to be identical. By forcing the data to be identical, all interpolation algorithms are guaranteed to properly reproduce an artifact-free loop. 7 The pdta-list Chunk 7.1 The HYDRA Data Structure The articulation data within a SoundFont 2 compatible file is contained in nine subchunks, named ÒhydraÓ after the mythical nine-headed beast. The structure has been designed for interchange purposes; it is not optimized for either run-time synthesis nor for on-the-fly editing. It is reasonable and proper for SoundFont compatible client programs to translate to and from the hydra structure as they read and write SoundFont compatible files. 7.2 The PHDR Subchunk The PHDR subchunk is a required subchunk listing all presets within the SoundFont compatible file. It is always a multiple of thirty eight bytes in length, and contains a minimum of two records, one record for each preset and one for a terminal record according to the structure: struct sfPresetHeader { CHAR achPresetName[20]; WORD wPreset; WORD wBank; WORD wPresetBagNdx; DWORD dwLibrary; DWORD dwGenre; DWORD dwMorphology; }; The ASCII character field achPresetName contains the name of the preset expressed in ASCII, with unused terminal characters filled with zero valued bytes. Preset names are case-sensitive. A unique name should always be assigned to each preset in the SoundFont compatible bank to enable identification. However, if a bank is read containing the erroneous state of presets with identical names, the presets should not be discarded. They should either be preserved as read or preferentially uniquely renamed. The WORD wPreset contains the MIDI Preset Number and the WORD wBank contains the MIDI Bank Number which apply to this preset. Note that the presets are not ordered within the SoundFont compatible bank. Presets should have a unique set of wPreset and wBank numbers. However, if two presets have identical values of both wPreset and wBank, the first occuring preset in the PHDR chunk is the active preset, but any others with the same wBank and wPreset values should be maintained so that they can be renumbered and used at a later time. The special case of a General MIDI percussion bank is handled conventionally by a wBank value of 128. If the value in either field is not a valid MIDI value of zero through 127, or 128 for wBank, the preset cannot be played but should be maintained. The WORD wPresetBagNdx is an index to the presetÕs layer list in the PBAG subchunk. Because the preset layer list is in the same order as the preset header list, the preset bag indicies will be monotonically increasing with increasing preset headers. The size of the PBAG subchunk in bytes will be equal to four times the terminal presetÕs wPresetBagNdx plus four. If the preset bag indicies are non-monotonic or if the terminal presetÕs wPresetBagNdx does not match the PBAG subchunk size, the file is structurally defective and should be rejected at load time. All presets except the terminal preset must have at least one layer; any preset with no layers should be ignored. The DWORDs dwLibrary; dwGenre and dwMorphology are reserved for future implementation in a preset library management function and should be preserved as read, and created as zero. The terminal sfPresetHeader record should never be accessed, and exists only to provide a terminal wPresetBagNdx with which to determine the number of layers in the last preset. All other values are conventionally zero, with the exception of achPresetName, which can optionally be ÒEOPÓ indicating end of presets. If the PHDR subchunk is missing, contains fewer than two records, or its size is not a multiple of 38 bytes, the file should be rejected as structurally unsound. 7.3 The PBAG Subchunk The PBAG subchunk is a required subchunk listing all preset layers within the SoundFont compatible file. It is always a multiple of four bytes in length, and contains one record for each preset layer plus one record for a terminal layer according to the structure: struct sfPresetBag { WORD wGenNdx; WORD wModNdx; }; The first layer in a given preset is located at that presetÕs wPresetBagNdx. The number of layers in the preset is determined by the difference between the next presetÕs wPresetBagNdx and the current wPresetBagNdx. The WORD wGenNdx is an index to the presetÕs layer list of generators in the PGEN subchunk, and the wModNdx is an index to its list of modulators in the PMOD subchunk. Because both the generator and modulator lists are in the same order as the preset header and layer lists, these indicies will be monotonically increasing with increasing preset layers. The size of the PMOD subchunk in bytes will be equal to ten times the terminal presetÕs wModNdx plus ten and the size of the PGEN subchunk in bytes will be equal to four times the terminal presetÕs wGenNdx plus four. If the generator or modulator indicies are non-monotonic or do not match the size of the respective PGEN or PMOD subchunks, the file is structurally defective and should be rejected at load time. If a preset has more than one layer, the first layer may be a global layer. A global layer is determined by the fact that the last generator in the list is not an Instrument generator. All generator lists must contain at least one generator with one exception - if a global layer exists for which there are no generators but only modulators. The modulator lists can contain zero or more modulators. If a layer other than the first layer lacks an Instrument generator as its last generator, that layer should be ignored. A global layer with no modulators and no generators should also be ignored. If the PBAG subchunk is missing, or its size is not a multiple of four bytes, the file should be rejected as structurally unsound. 7.4 The PMOD Subchunk The PMOD subchunk is a required subchunk listing all preset layer modulators within the SoundFont compatible file. It is always a multiple of ten bytes in length, and contains zero or more modulators plus a terminal record according to the structure: struct sfModList { SFModulator sfModSrcOper; SFGenerator sfModDestOper; SHORT modAmount; SFModulator sfModAmtSrcOper; SFTransform sfModTransOper; }; The preset layerÕs wModNdx points to the first modulator for that preset layer, and the number of modulators present for a preset layer is determined by the difference between the next higher preset layerÕs wModNdx and the current presetÕs wModNdx. A difference of zero indicates there are no modulators in this preset layer. The sfModSrcOper is a value of one of the SFModulator enumeration type values. Unknown or undefined values are ignored. This value indicates the source of data for the modulator. The sfModDestOper is a value of one of the SFGenerator enumeration type values. Unknown or undefined values are ignored. This value indicates the destination of the modulator. The SHORT modAmount is a signed value indicating the degree to which the source modulates the destination. A zero value indicates there is no fixed amount. The sfModAmtSrcOper is a value of one of the SFModulator enumeration type values. Unknown or undefined values are ignored. This value indicates the degree to which the source modulates the destination is to be controlled by the specified modulation source. The sfModTransOper is a value of one of the SFTransform enumeration type values. Unknown or undefined values are ignored. This value indicates that a transform of the specified type will be applied to the modulation source before application to the modulator. The terminal record conventionally contains zero in all fields, and is always ignored. A modulator is defined by its sfModSrcOper, its sfModDestOper, and its sfModSrcAmtOper. All modulators within a layer must have a unique set of these three enumerators. If a second modulator is encountered with the same three enumerators as a previous modulator with the same layer, the first modulator will be ignored. Modulators in the PMOD subchunk act as additively relative modulators with respect to those in the IMOD subchunk. In other words, a PMOD modulator can increase or decrease the amount of an IMOD modulator. In SoundFont 2.00, no modulators have yet been defined, and the PMOD subchunk will always consist of ten zero valued bytes. If the PMOD subchunk is missing, or its size is not a multiple of ten bytes, the file should be rejected as structurally unsound. 7.5 The PGEN Subchunk The PGEN chunk is a required chunk containing a list of preset layer generators for each preset layer within the SoundFont compatible file. It is always a multiple of four bytes in length, and contains one or more generators for each preset layer (except a global layer containing only modulators) plus a terminal record according to the structure: struct sfGenList { SFGenerator sfGenOper; genAmountType genAmount; }; where the types are defined: typedef struct { BYTE byLo; BYTE byHi; } rangesType; typedef union { rangesType ranges; SHORT shAmount; WORD wAmount; } genAmountType; The sfGenOper is a value of one of the SFGenerator enumeration type values. Unknown or undefined values are ignored. This value indicates the type of generator being indicated. The genAmount is the value to be assigned to the specified generator. Note that this can be of three formats. Certain generators specify a range of MIDI key numbers of MIDI velocities, with a minimum and maximum value. Other generators specify an unsigned WORD value. Most generators, however, specify a signed 16 bit SHORT value. The preset layerÕs wGenNdx points to the first generator for that preset layer. Unless the layer is a global layer, the last generator in the list is an ÒInstrumentÓ generator, whose value is a pointer to the instrument associated with that layer. If a Òkey rangeÓ generator exists for the preset layer, it is always the first generator in the list for that preset layer. If a Òvelocity rangeÓ generator exists for the preset layer, it will only be preceded by a key range generator. If any generators follow an Instrument generator, they will be ignored. A generator is defined by its sfGenOper. All generators within a layer must have a unique sfGenOper enumerator. If a second generator is encountered with the same sfGenOper enumerator as a previous generator with the same layer, the first generator will be ignored. Generators in the PGEN subchunk are applied relative to generators in the IGEN subchunk in an additve manner. In other words, PGEN generators increase or decrease the value of an IGEN generator. If the PGEN subchunk is missing, or its size is not a multiple of four bytes, the file should be rejected as structurally unsound. If a key range generator is present and not the first generator, it should be ignored. If a velocity range generator is present, and is preceded by a generator other than a key range generator, it should be ignored. If a non-global list does not end in an instrument generator, layer should be ignored. If the instrument generator value is equal to or greater than the terminal instrument, the file should be rejected as structurally unsound. 7.6 The INST Subchunk The inst subchunk is a required subchunk listing all instruments within the SoundFont compatible file. It is always a multiple of twenty two bytes in length, and contains a minimum of two records, one record for each instrument and one for a terminal record according to the structure: struct sfInst { CHAR achInstName[20]; WORD wInstBagNdx; }; The ASCII character field achInstName contains the name of the instrument expressed in ASCII, with unused terminal characters filled with zero valued bytes. Instrument names are case-sensitive. A unique name should always be assigned to each instrument in the SoundFont compatible bank to enable identification. However, if a bank is read containing the erroneous state of instruments with identical names, the instruments should not be discarded. They should either be preserved as read or preferentially uniquely renamed. The WORD wInstBagNdx is an index to the instrumentÕs split list in the IBAG subchunk. Because the instrument split list is in the same order as the instrument list, the instrument bag indicies will be monotonically increasing with increasing instruments. The size of the IBAG subchunk in bytes will be four greater than four times the terminal (EOI) instrumentÕs wInstBagNdx. If the instrument bag indicies are non-monotonic or if the terminal instrumentÕs wInstBagNdx does not match the IBAG subchunk size, the file is structurally defective and should be rejected at load time. All instruments except the terminal instrument must have at least one split; any preset with no splits should be ignored. The terminal sfInst record should never be accessed, and exists only to provide a terminal wInstBagNdx with which to determine the number of splits in the last instrument. All other values are conventionally zero, with the exception of achInstName, which should be ÒEOIÓ indicating end of instruments. If the INST subchunk is missing, contains fewer than two records, or its size is not a multiple of 22 bytes, the file should be rejected as structurally unsound. All instruments present in the inst subchunk are typically referenced by a preset layer. However, a file containing any ÒorphanedÓ instruments need not be rejected. SoundFont compatible applications can optionally ignore or filter out these orphaned instruments based on user preference. 7.7 The IBAG Subchunk The IBAG subchunk is a required subchunk listing all instrument splits within the SoundFont compatible file. It is always a multiple of four bytes in length, and contains one record for each instrument split plus one record for a terminal layer according to the structure: struct sfInstBag { WORD wInstGenNdx; WORD wInstModNdx; }; The first split in a given instrument is located at that instrumentÕs wInstBagNdx. The number of splits in the instrument is determined by the difference between the next instrumentÕs wInstBagNdx and the current wInstBagNdx. The WORD wInstGenNdx is an index to the instrument splitÕs list of generators in the IGEN subchunk, and the wInstModNdx is an index to its list of modulators in the IMOD subchunk. Because both the generator and modulator lists are in the same order as the instrument and split lists, these indicies will be monotonically increasing with increasing splits. The size of the IMOD subchunk in bytes will be equal to ten times the terminal instrumentÕs wModNdx plus ten and the size of the IGEN subchunk in bytes will be equal to four times the terminal instrumentÕs wGenNdx plus four. If the generator or modulator indicies are non-monotonic or do not match the size of the respective IGEN or IMOD subchunks, the file is structurally defective and should be rejected at load time. If an instrument has more than one split, the first split may be a global split. A global split is determined by the fact that the last generator in the list is not a sampleID generator. All generator lists must contain at least one generator with one exception - if a global split exists for which there are no generators but only modulators. The modulator lists can contain zero or more modulators. If a split other than the first split lacks a sampleID generator as its last generator, that split should be ignored. A global split with no modulators and no generators should also be ignored. If the IBAG subchunk is missing, or its size is not a multiple of four bytes, the file should be rejected as structurally unsound. 7.8 The IMOD Subchunk The IMOD subchunk is a required subchunk listing all instrument split modulators within the SoundFont compatible file. It is always a multiple of ten bytes in length, and contains zero or more modulators plus a terminal record according to the structure: struct sfModList { SFModulator sfModSrcOper; SFGenerator sfModDestOper; SHORT modAmount; SFModulator sfModAmtSrcOper; SFTransform sfModTransOper; }; The splitÕs wInstModNdx points to the first modulator for that split, and the number of modulators present for a split is determined by the difference between the next higher splitÕs wInstModNdx and the current splitÕs wModNdx. A difference of zero indicates there are no modulators in this split. The sfModSrcOper is a value of one of the SFModulator enumeration type values. Unknown or undefined values are ignored. This value indicates the source of data for the modulator. The sfModDestOper is a value of one of the SFGenerator enumeration type values. Unknown or undefined values are ignored. This value indicates the destination of the modulator. The SHORT modAmount is a signed value indicating the degree to which the source modulates the destination. A zero value indicates there is no fixed amount. The sfModAmtSrcOper is a value of one of the SFModulator enumeration type values. Unknown or undefined values are ignored. This value indicates the degree to which the source modulates the destination is to be controlled by the specified modulation source. The sfModTransOper is a value of one of the SFTransform enumeration type values. Unknown or undefined values are ignored. This value indicates that a transform of the specified type will be applied to the modulation source before application to the modulator. The terminal record conventionally contains zero in all fields, and is always ignored. A modulator is defined by its sfModSrcOper, its sfModDestOper, and its sfModSrcAmtOper. All modulators within a split must have a unique set of these three enumerators. If a second modulator is encountered with the same three enumerators as a previous modulator with the same split, the first modulator will be ignored. Modulators in the IMOD subchunk are absolute. This means that an IMOD modulator replaces, rather than adding to, a default modulator. In SoundFont 2.00, no modulators have yet been defined, and the IMOD subchunk will always consist of ten zero valued bytes. If the IMOD subchunk is missing, or its size is not a multiple of ten bytes, the file should be rejected as structurally unsound. 7.9 The IGEN Subchunk The IGEN chunk is a required chunk containing a list of split generators for each instrument split within the SoundFont compatible file. It is always a multiple of four bytes in length, and contains one or more generators for each split (except a globalsplit containing only modulators) plus a terminal record according to the structure: struct sfInstGenList { SFGenerator sfGenOper; genAmountType genAmount; }; where the types are defined as in the PGEN layer above. The genAmount is the value to be assigned to the specified generator. Note that this can be of three formats. Certain generators specify a range of MIDI key numbers of MIDI velocities, with a minimum and maximum value. Other generators specify an unsigned WORD value. Most generators, however, specify a signed 16 bit SHORT value. The splitÕs wInstGenNdx points to the first generator for that split. Unless the split is a global split, the last generator in the list is a ÒsampleIDÓ generator, whose value is a pointer to the sample associated with that split. If a Òkey rangeÓ generator exists for the split, it is always the first generator in the list for that split. If a Òvelocity rangeÓ generator exists for the split, it will only be preceded by a key range generator. If any generators follow a sampleID generator, they will be ignored. A generator is defined by its sfGenOper. All generators within a split must have a unique sfGenOper enumerator. If a second generator is encountered with the same sfGenOper enumerator as a previous generator with the same split, the first generator will be ignored. Generators in the IGEN subchunk are absolute in nature. This means that an IGEN generator replaces, rather than adding to, the default value for the generator. If the IGEN subchunk is missing, or its size is not a multiple of four bytes, the file should be rejected as structurally unsound. If a key range generator is present and not the first generator, it should be ignored. If a velocity range generator is present, and is preceded by a generator other than a key range generator, it should be ignored. If a non-global list does not end in a sampleID generator, the split should be ignored. If the sampleID generator value is equal to or greater than the terminal sampleID, the file should be rejected as structurally unsound. 7.10 The SHDR Subchunk The SHDR chunk is a required subchunk listing all samples within the smpl subchunk and any referenced ROM samples. It is always a multiple of forty six bytes in length, and contains one record for each sample plus a terminal record according to the structure: struct sfSample { CHAR achSampleName[20]; DWORD dwStart; DWORD dwEnd; DWORD dwStartloop; DWORD dwEndloop; DWORD dwSampleRate; BYTE byOriginalPitch; CHAR chPitchCorrection; WORD wSampleLink; SFSampleLink sfSampleType; }; The ASCII character field achSampleName contains the name of the sample expressed in ASCII, with unused terminal characters filled with zero valued bytes. Sample names are case-sensitive. A unique name should always be assigned to each sample in the SoundFont compatible bank to enable identification. However, if a bank is read containing the erroneous state of samples with identical names, the samples should not be discarded. They should either be preserved as read or preferentially uniquely renamed. The DWORD dwStart contains the index, in sample data points, from the beginning of the sample data field to the first data point of this sample. The DWORD dwEnd contains the index, in sample data points, from the beginning of the sample data field to the first of the set of 46 zero valued data points following this sample. The DWORD dwStartloop contains the index, in sample data points, from the beginning of the sample data field to the first datapoint in the loop of this sample. The DWORD dwEndloop contains the index, in sample data points, from the beginning of the sample data field to the first datapoint following the loop of this sample. Note that this is the data point Òequivalent toÓ the first loop datapoint, and that to produce portable artifact free loops, the eight proximal datapoints surrounding both the Startloop and Endloop points should be identical. The values of dwStart, dwEnd, dwStartloop, and dwEndloop must all be within the range of the sample data field included in the SoundFont compatible bank or referenced in the sound ROM. Also, to allow a variety of hardware platforms to be able to reproduce the data, the samples have a minimum length of 48 data points, a minimum loop size of 32 data points, and a minimum of 8 valid points prior to dwStartloop and after dwEndloop. Thus dwStart must be less than dwStartloop-7, dwStartloop must be less than dwEndloop-31, and dwEndloop must be less than dwEnd-7. If these constraints are not met, the sound may optionally not be played if the hardware cannot support artifact-free playback for the parameters given. The DWORD dwSampleRate contains the sample rate, in Hertz, at which this sample was acquired or to which it was most recently converted. Values of greater than 50000 or less than 400 may not be reproducable by some hardware platforms and should be avoided. A value of zero is illegal. If an illegal or impractical value is encountered, the nearest practical value should be used. The BYTE byOriginalPitch contains the MIDI key number of the recorded pitch of the sample. For example, a recording of an instrument playing middle C (261.62 Hz) should receive a value of 60. This value is used as the default Òroot keyÓ for the sample, so that in the example, a MIDI key-on command for note number 60 would reproduce the sound at its original pitch. For unpitched sounds, a conventional value of 255 should be used. Values between 128 and 254 are illegal. Whenever an illegal value or a value of 255 is encountered, the value 60 should be used. The CHAR chPitchCorrection contains a pitch correction in cents which should be applied to the sample on playback. The purpose of this field is to compensate for any pitch errors during the sample recording process. The correction value is that of the correction to be applied. For example, if the sound is 4 cents sharp, a correction bringing it 4 cents flat is required, thus the value should be -4. The value in sfSampleType is an enumeration with eight defined values: monoSample = 1, rightSample = 2, leftSample = 4, linkedSample = 8, RomMonoSample = 32769, RomRightSample = 32770, RomLeftSample = 32772, and RomLinkedSample = 32776. It can be seen that this is encoded such that bit 15 of the 16 bit value is set if the sample is in ROM, and reset if it is included in the SoundFont compatible bank. The four LS bits of the word are then exclusively set indicating mono, left, right, or linked. If the sound is flagged as a ROM sample and no valid ÒiromÓ subchunk is included, the file is structurally defective and should be rejected at load time. If sfSampleType indicates a mono sample, then wSampleLink is undefined and its value should be conventionally zero, but will be ignored regardless of value. If sfSampleType indicates a left or right sample, then wSampleLink is the sample header index of the associated right or left stereo sample respectively. Both samples should be played entirely syncrhonously, with their pitch controlled by the right sampleÕs generators. All non-pitch generators should apply as normal; in particular the panning of the individual samples to left and right should be accomplished via the pan generator. Left-right pairs should always be found within the same instrument. Note also that no instrument should be designed in which it is possible to activate more than one instance of a particular stereo pair. The linked sample type is not currently fully defined in the SoundFont 2 specification, but will ultimately support a circularly linked list of samples using wSampleLink. The terminal sample record is never referenced, and is conventionally entirely zero with the exception of achSampleName, which should be ÒEOSÓ indicating end of samples. All samples present in the smpl subchunk are typically referenced by an instrument, however a file containing any ÒorphanedÓ samples need not be rejected. SoundFont compatible applications can optionally ignore or filter out these orphaned samples according to user preference. If the SHDR subchunk is missing, or its is size is not a multiple of 46 bytes the file should be rejected as structurally unsound. 8 Enumerators 8.1 Generator Enumerators 8.1.1 Kinds of Generator Enumerators Five kinds of Generator Enumerators exist: Index Generators, Range Generators, Substitution Generators, Sample Generators, and Value Generators. An Index GeneratorÕs amount is an index into another data structure. The only two Index Generators are Instrument and sampleID. A Range Generator defines a range of note-on parameters outside of which the layer or split is undefined. Two Range Generators are currently defined, keyRange and velRange. Substitution Generators are generators which substitute a value for a note-on parameter. Two Substitution Generators are currently defined, overridingKeyNumber and overridingVelocity. Sample Generators are generators which directly affect a sampleÕs properties. These generators are undefined at the layer level. The currently defined Sample Generators are the eight address offset generators, the sampleModes generator, the Overriding Root Key generator and the Exclusive Class generator. Value Generators are generators whose value directly affects a signal processing parameter. Most generators are value generators. 8.1.2 Generator Enumerators Defined The following is an exhaustive list of SoundFont 2.00 generators and their strict definitions: 0 startAddrsOffset The offset, in sample data points, beyond the Start sample header parameter to the first sample data point to be played for this instrument. For example, if Start were 7 and startAddrOffset were 2, the first sample data point played would be sample data point 9. 1 endAddrsOffset The offset, in sample sample data points, beyond the End sample header parameter to the last sample data point to be played for this instrument. For example, if End were 17 and endAddrOffset were -2, the last sample data point played would be sample data point 15. 2 startloopAddrsOffset The offset, in sample data points, beyond the Startloop sample header parameter to the first sample data point to be repeated in the loop for this instrument. For example, if Startloop were 10 and startloopAddrOffset were -1, the first repeated loop sample data point would be sample data point 9. 3 endloopAddrsOffset The offset, in sample data points, beyond the Endloop sample header parameter to the sample data point considered equivalent to the Startloop sample data point for the loop for this instrument. For example, if Endloop were 15 and endloopAddrOffset were 2, sample data point 17 would be considered equivalent to the Startloop sample data point, and hence sample data point 16 would effectively precede Startloop during looping. 4 startAddrsCoarseOffset The offset, in 32768 sample data point increments beyond the Start sample header parameter and the first sample data point to be played in this instrument. This parameter is added to the startAddrsOffset parameter. For example, if Start were 5, startAddrOffset were 3 and startAddrCoarseOffset were 2, the first sample data point played would be sample data point 65544. 5 modLfoToPitch This is the degree, in cents, to which a full scale excursion of the Modulation LFO will influence pitch. A positive value indicates a positive LFO excursion increases pitch; a negative value indicates a positive excursion decreases pitch. Pitch is always modified logarithmically, that is the deviation is in cents, semitones, and octaves rather than in Hz. For example, a value of 100 indicates that the pitch will first rise 1 semitone, then fall one semitone. 6 vibLfoToPitch This is the degree, in cents, to which a full scale excursion of the Vibrato LFO will influence pitch. A positive value indicates a positive LFO excursion increases pitch; a negative value indicates a positive excursion decreases pitch. Pitch is always modified logarithmically, that is the deviation is in cents, semitones, and octaves rather than in Hz. For example, a value of 100 indicates that the pitch will first rise 1 semitone, then fall one semitone. 7 modEnvToPitch This is the degree, in cents, to which a full scale excursion of the Modulation Envelope will influence pitch. A positive value indicates an increase in pitch; a negative value indicates a decrease in pitch. Pitch is always modified logarithmically, that is the deviation is in cents, semitones, and octaves rather than in Hz. For example, a value of 100 indicates that the pitch will rise 1 semitone at the envelope peak. 8 initialFilterFc This is the cutoff and resonant frequency of the lowpass filter in absolute cent units. The lowpass filter is defined as a second order resonant pole pair whose pole frequency in Hz is defined by the Initial Filter Cutoff parameter. When the cutoff frequency exceeds 20kHz and the Q (resonance) of the filter is zero, the filter does not affect the signal. 9 initialFilterQ This is the height above DC gain in centibels which the filter resonance exhibits at the cutoff frequency. A value of zero or less indicates the filter is not resonant; the gain at the cutoff frequency (pole angle) may be less than zero when zero is specified. The filter gain at DC is also affected by this parameter such that the gain at DC is reduced by half the specified gain. For example, for a value of 100, the filter gain at DC would be 5 dB below unity gain, and the height of the resonant peak would be 10 dB above the DC gain, or 5 dB above unity gain. Note also that if initialFilterQ is set to zero or less and the cutoff frequency exceeds 20 kHz, then the filter response is flat and unity gain. 10 modLfoToFilterFc This is the degree, in cents, to which a full scale excursion of the Modulation LFO will influence filter cutoff frequency. A positive number indicates a positive LFO excursion increases cutoff frequency; a negative number indicates a positive excursion decreases cutoff frequency. Filter cutoff frequency is always modified logarithmically, that is the deviation is in cents, semitones, and octaves rather than in Hz. For example, a value of 1200 indicates that the cutoff frequency will first rise 1 octave, then fall one octave. 11 modEnvToFilterFc This is the degree, in cents, to which a full scale excursion of the Modulation Envelope will influence filter cutoff. A positive number indicates an increase in cutoff frequency; a negative number indicates a decrease in filter cutoff. Filter cutoff is always modified logarithmically, that is the deviation is in cents, semitones, and octaves rather than in Hz. For example, a value of 1000 indicates that the cutoff frequency will rise one octave at the envelope attack peak. 12 endAddrsCoarseOffset The offset, in 32768 sample data point increments beyond the End sample header parameter and the last sample data point to be played in this instrument. This parameter is added to the endAddrsOffset parameter. For example, if End were 65536, startAddrOffset were -3 and startAddrCoarseOffset were -1, the last sample data point played would be sample data point 32765. 13 modLfoToVolume This is the degree, in centibels, to which a full scale excursion of the Modulation LFO will influence volume. A positive number indicates a positive LFO excursion increases volume; a negative number indicates a positive excursion decreases volume. Volume is always modified logarithmically, that is the deviation is in decibels rather than in linear amplitude. For example, a value of 100 indicates that the volume will first rise ten dB, then fall ten dB. 14 unused1 Unused, reserved. Should be ignored if encountered. 15 chorusEffectsSend This is the degree, in 0.1% units, to which the audio output of the note is sent to the chorus effects processor. A value of 0% or less indicates no signal is sent from this note; a value of 100% or more indicates the note is sent at full level. Note that this parameter has no effect on the amount of this signal sent to the ÒdryÓ or unprocessed portion of the output. For example, a value of 250 indicates that the signal is sent at 25% of full level (attenuation of 12 dB from full level) to the chorus effects processor. 16 reverbEffectsSend This is the degree, in 0.1% units, to which the audio output of the note is sent to the reverb effects processor. A value of 0% or less indicates no signal is sent from this note; a value of 100% or more indicates the note is sent at full level. Note that this parameter has no effect on the amount of this signal sent to the ÒdryÓ or unprocessed portion of the output. For example, a value of 250 indicates that the signal is sent at 25% of full level (attenuation of 12 dB from full level) to the reverb effects processor. 17 pan This is the degree, in 0.1% units, to which the ÒdryÓ audio output of the note is positioned to the left or right output. A value of -50% or less indicates the signal is sent entirely to the left output and not sent to the right output; a value of +50% or more indicates the note is sent entirely to the right and not sent to the left. A value of zero places the signal centered between left and right. For example, a value of -250 indicates that the signal is sent at 75% of full level to the left output and 25% of full level to the right output. 18 unused2 Unused, reserved. Should be ignored if encountered. 19 unused3 Unused, reserved. Should be ignored if encountered. 20 unused4 Unused, reserved. Should be ignored if encountered. 21 delayModLFO This is the delay time, in absolute timecents, from key on until the Modulation LFO begins its upward ramp from zero value. A value of 0 indicates a 1 second delay. A negative value indicates a delay less than one second and a positive value a delay longer than one second. The most negative number (-32768) conventionally indicates no delay. For example, a delay of 10 msec would be 1200log2(.01) = -7973. 22 freqModLFO This is the frequency, in absolute cents, of the Modulation LFOÕs triangular period. A value of zero indicates a frequency of 8.176 Hz. A negative value indicates a frequency less than 8.176 Hz; a positive value a frequency greater than 8.176 Hz. For example, a frequency of 10 mHz would be 1200log2(.01/8.176) = -11610. 23 delayVibLFO This is the delay time, in absolute timecents, from key on until the Vibrato LFO begins its upward ramp from zero value. A value of 0 indicates a 1 second delay. A negative value indicates a delay less than one second; a positive value a delay longer than one second. The most negative number (-32768) conventionally indicates no delay. For example, a delay of 10 msec would be 1200log2(.01) = -7973. 24 freqVibLFO This is the frequency, in absolute cents, of the Vibrato LFOÕs triangular period. A value of zero indicates a frequency of 8.176 Hz. A negative value indicates a frequency less than 8.176 Hz; a positive value a frequency greater than 8.176 Hz. For example, a frequency of 10 mHz would be 1200log2(.01/8.176) = -11610. 25 delayModEnv This is the delay time, in absolute timecents, between key on and the start of the attack phase of the Modulation envelope. A value of 0 indicates a 1 second delay. A negative value indicates a delay less than one second; a positive value a delay longer than one second. The most negative number (-32768) conventionally indicates no delay. For example, a delay of 10 msec would be 1200log2(.01) = -7973. 26 attackModEnv This is the time, in absolute timecents, from the end of the Modulation Envelope Delay Time until the point at which the Modulation Envelope value reaches its peak. Note that the attack is ÒconvexÓ; the curve is nominally such that when applied to a decibel or semitone parameter, the result is linear in amplitude or Hz respectively. A value of 0 indicates a 1 second attack time. A negative value indicates a time less than one second; a positive value a time longer than one second. The most negative number (-32768) conventionally indicates instantaneous attack. For example, an attack time of 10 msec would be 1200log2(.01) = -7973. 27 holdModEnv This is the time, in absolute timecents, from the end of the attack phase to the entry into decay phase, during which the envelope value is held at its peak. A value of 0 indicates a 1 second hold time. A negative value indicates a time less than one second; a positive value a time longer than one second. The most negative number (-32768) conventionally indicates no hold phase. For example, a hold time of 10 msec would be 1200log2(.01) = -7973. 28 decayModEnv This is the time, in absolute timecents, for a 100% change in the Modulation Envelope value during decay phase. For the Modulation Envelope, the decay phase linearly ramps toward the sustain level. If the sustain level were zero, the Modulation Envelope Decay Time would be the time spent in decay phase. A value of 0 indicates a 1 second decay time for a zero sustain level. A negative value indicates a time less than one second; a positive value a time longer than one second. For example, a decay time of 10 msec would be 1200log2(.01) = -7973. 29 sustainModEnv This is the decrease in level, expressed in 0.1% units, over which the Modulation Envelope value ramps during the decay phase. For the Modulation Envelope, the sustain level is properly expressed in percent of full scale. Because the volume envelope sustain level is expressed as an attenuation from full scale, the sustain level is analogously expressed as a decrease from full scale. A value of 0 indicates the sustain level is full level; this implies a zero duration of decay phase regardless of decay time. A positive value indicates a decay to the corresponding level. Values less than zero are to be interpreted as zero; values above 1000 are to be interpreted as 1000. For example, a sustain level which corresponds to an absolute value 40% of peak would be 600. 30 releaseModEnv This is the time, in absolute timecents, for a 100% change in the Modulation Envelope value during release phase. For the Modulation Envelope, the release phase linearly ramps toward zero from the current level. If the current level were full scale, the Modulation Envelope Release Time would be the time spent in release phase until zero value were reached. A value of 0 indicates a 1 second decay time for a release from full level. A negative value indicates a time less than one second; a positive value a time longer than one second. For example, a release time of 10 msec would be 1200log2(.01) = -7973. 31 keynumToModEnvHold This is the degree, in timecent per keynumber units, to which the hold time of the Modulation Envelope is decreased by increasing MIDI key number. The hold time at key number 60 is always unchanged. The unit scaling is such that a value of 100 provides a hold time which tracks the keyboard; that is, an upward octave causes the hold time to halve. For example, if the Modulation Envelope Hold Time were - 7973 = 10 msec and the Key Number to Mod Env Hold were 50 when key number 36 was played, the hold time would be 20 msec. 32 keynumToModEnvDecay This is the degree, in timecent per keynumber units, to which the hold time of the Modulation Envelope is decreased by increasing MIDI key number. The hold time at key number 60 is always unchanged. The unit scaling is such that a value of 100 provides a hold time which tracks the keyboard; that is, an upward octave causes the hold time to halve. For example, if the Modulation Envelope Hold Time were - 7973 = 10 msec and the Key Number to Mod Env Hold were 50 when key number 36 was played, the hold time would be 20 msec. 33 delayVolEnv This is the delay time, in absolute timecents, between key on and the start of the attack phase of the Volume envelope. A value of 0 indicates a 1 second delay. A negative value indicates a delay less than one second; a positive value a delay longer than one second. The most negative number (-32768) conventionally indicates no delay. For example, a delay of 10 msec would be 1200log2(.01) = -7973. 34 attackVolEnv This is the time, in absolute timecents, from the end of the Volume Envelope Delay Time until the point at which the Volume Envelope value reaches its peak. Note that the attack is ÒconvexÓ; the curve is nominally such that when applied to the decibel volume parameter, the result is linear in amplitude. A value of 0 indicates a 1 second attack time. A negative value indicates a time less than one second; a positive value a time longer than one second. The most negative number (-32768) conventionally indicates instantaneous attack. For example, an attack time of 10 msec would be 1200log2(.01) = -7973. 35 holdVolEnv This is the time, in absolute timecents, from the end of the attack phase to the entry into decay phase, during which the Volume envelope value is held at its peak. A value of 0 indicates a 1 second hold time. A negative value indicates a time less than one second; a positive value a time longer than one second. The most negative number (-32768) conventionally indicates no hold phase. For example, a hold time of 10 msec would be 1200log2(.01) = -7973. 36 decayVolEnv This is the time, in absolute timecents, for a 100% change in the Volume Envelope value during decay phase. For the Volume Envelope, the decay phase linearly ramps toward the sustain level, causing a constant dB change for each time unit. If the sustain level were -100dB, the Volume Envelope Decay Time would be the time spent in decay phase. A value of 0 indicates a 1 second decay time for a zero sustain level. A negative value indicates a time less than one second; a positive value a time longer than one second. For example, a decay time of 10 msec would be 1200log2(.01) = -7973. 37 sustainVolEnv This is the decrease in level, expressed in centibels, over which the Volume Envelope value ramps during the decay phase. For the Volume Envelope, the sustain level is best expressed in cB of attenuation from full scale. A value of 0 indicates the sustain level is full level; this implies a zero duration of decay phase regardless of decay time. A positive value indicates a decay to the corresponding level. Values less than zero are to be interpreted as zero; conventionally 1000 indicates full attenuation. For example, a sustain level which corresponds to an absolute value 12dB below of peak would be 120. 38 releaseVolEnv This is the time, in absolute timecents, for a 100% change in the Volume Envelope value during release phase. For the Volume Envelope, the release phase linearly ramps toward zero from the current level, causing a constant dB change for each time unit. If the current level were full scale, the Volume Envelope Release Time would be the time spent in release phase until -100dB attenuation were reached. A value of 0 indicates a 1 second decay time for a release from full level. A negative value indicates a time less than one second; a positive value a time longer than one second. For example, a release time of 10 msec would be 1200log2(.01) = -7973. 39 keynumToVolEnvHold This is the degree, in timecent per keynumber units, to which the hold time of the Volume Envelope is decreased by increasing MIDI key number. The hold time at key number 60 is always unchanged. The unit scaling is such that a value of 100 provides a hold time which tracks the keyboard; that is, an upward octave causes the hold time to halve. For example, if the Volume Envelope Hold Time were -7973 = 10 msec and the Key Number to Vol Env Hold were 50 when key number 36 was played, the hold time would be 20 msec. 40 keynumToVolEnvDecay This is the degree, in timecent per keynumber units, to which the hold time of the Volume Envelope is decreased by increasing MIDI key number. The hold time at key number 60 is always unchanged. The unit scaling is such that a value of 100 provides a hold time which tracks the keyboard; that is, an upward octave causes the hold time to halve. For example, if the Volume Envelope Hold Time were -7973 = 10 msec and the Key Number to Vol Env Hold were 50 when key number 36 was played, the hold time would be 20 msec. 41 instrument This is the index into the INST subchunk providing the instrument to be used for the current layer. A value of zero indicates the first instrument in the list. The value should never exceed two less than the size of the instrument list. The instrument enumerator is the terminal generator for PGEN layers. As such, it should only appear in the PGEN subchunk, and it must appear as the last generator enumerator in all but the global layer. 42 reserved1 Unused, reserved. Should be ignored if encountered. 43 keyRange This is the minimum and maximum MIDI key number values for which this preset, layer, instrument or split is active. The LS byte indicates the highest and the MS byte the lowest valid key . The keyRange enumerator is optional, but when it does appear, it must be the first generator in the preset layer or instrument split generator list. 44 velRange This is the minimum and maximum MIDI velocity values for which this preset, layer, instrument or split is active. The LS byte indicates the highest and the MS byte the lowest valid velocity . The velRange enumerator is optional, but when it does appear, it must be preceded only by keyRange in the preset layer or instrument split generator list. 45 startloopAddrsCoarseOffset The offset, in 32768 sample data point increments beyond the Startloop sample header parameter and the first sample data point to be repeated in this instrumentÕs loop. This parameter is added to the startloopAddrsOffset parameter. For example, if Startloop were 5, startloopAddrOffset were 3 and startAddrCoarseOffset were 2, the first sample data point in the loop would be sample data point 65544. 46 keynum This enumerator forces the MIDI key number to effectively be interpreted as the value given. This generator caon only appear at the instrument level. Valid values are from 0 to 127. 47 velocity This enumerator forces the MIDI velocity to effectively be interpreted as the value given. This generator caon only appear at the instrument level. Valid values are from 0 to 127. 48 initialAttenuation This is the attenuation, in centibels, by which a note is attenuated below full scale. A value of zero indicates no attenuation; the note will be played at full scale. For example, a value of 60 indicates the note will be played at 6 dB below full scale for the note. 49 reserved2 Unused, reserved. Should be ignored if encountered. 50 endloopAddrsCoarseOffset The offset, in 32768 sample data point increments beyond the Endloop sample header parameter parameter to the sample data point considered equivalent to the Startloop sample data point for the loop for this instrument. This parameter is added to the endloopAddrsOffset parameter. For example, if Endloop were 5, endloopAddrOffset were 3 and endAddrCoarseOffset were 2, sample data point 65544 would be considered equivalent to the Startloop sample data point, and hence sample data point 65543 would effectively precede Startloop during looping. 51 coarseTune This is a pitch offset, in semitones, which should be applied to the note. A positive value indicates the sound is reproduced at a higher pitch; a negative value indicates a lower pitch. For example, a Coarse Tune value of -4 would cause the sound to be reproduced four semitones flat. 52 fineTune This is a pitch offset, in cents, which should be applied to the note. It is additive with coarseTune. A positive value indicates the sound is reproduced at a higher pitch; a negative value indicates a lower pitch. . For example, a Fine Tuning value of -5 would cause the sound to be reproduced five cents flat. 53 sampleID This is the index into the SHDR subchunk providing the sample to be used for the current split. A value of zero indicates the first sample in the list. The value should never exceed two less than the size of the sample list. The sampleID enumerator is the terminal generator for IGEN splits. As such, it should only appear in the IGEN subchunk, and it must appear as the last generator enumerator in all but the global split. 54 sampleModes This enumerator indicates a value which gives a variety of Boolean flags describing the sample for the current instrument split. The sampleModes should only appear in the IGEN subchunk, and should not appear in the global split. The two LS bits of the value indicate the type of loop in the sample: 0 indicates a sound reproduced with no loop, 1 indicates a sound which loops continuously, 2 is unused but should be interpreted as indicating no loop, and 3 indicates a sound which loops for the duration of key depression then proceeds to play the remainder of the sample. 55 reserved3 Unused, reserved. Should be ignored if encountered. 56 scaleTuning This parameter represents the degree to which MIDI key number influences pitch. A value of zero indicates that MIDI key number has no effect on pitch; a value of 100 represents the usual tempered semitone scale. 57 exclusiveClass This parameter provides the capability for a key depression in a given instrument to terminate the playback of other instruments. This is particularly useful for percussive instruments such as a hihat cymbal. An exclusive class value of zero indicates no exclusive class; no special action is taken. Any other value indicates that when this note is initiated, any other sounding note with the same exclusive class value should be rapidly terminated. The exclusive class generator can only appear at the instrument level. The scope of the exclusive class is the entire preset. In other words, any other instrument split within the same preset holding a corresponding exclusive class will be terminated. 58 overridingRootKey This parameter represents the MIDI key number at which the sample is to be played back at its original sample rate. If not present, or if present with a value of -1, then the sample header parameter Original Key is used in its place. If it is present in the range 0-127, then the indicated key number will cause the sample to be played back at its sample header Sample Rate. For example, if the sample were a recording of a piano middle C (Original Key = 60) at a sample rate of 22.050 kHz, and Root Key were set to 69, then playing MIDI key number 69 (A above middle C) would cause a piano note of pitch middle C to be heard. 59 unused5 Unused, reserved. Should be ignored if encountered. 60 endOper Unused, reserved. Should be ignored if encountered. Unique name provides value to end of defined list. 8.1.3 Generator Summary The following tables give the ranges and default values for all SoundFont 2.00 defined generators. # Name Unit Abs Zero Min Useful Max Useful Default Value 0 startAddrsOffset+ smpls 0 0 None * * 0 None 1 endAddrsOffset+ smpls 0 * * 0 None 0 None 2 startloopAddrsOffset+ smpls 0 * * * * 0 None 3 endloopAddrsOffset+ smpls 0 * * * * 0 None 4 startAddrsCoarseOffset+ 32k smpls 0 0 None * * 0 None 5 modLfoToPitch cent fs 0 -12000 -10 oct 12000 10 oct 0 None 6 vibLfoToPitch cent fs 0 -12000 -10 oct 12000 10 oct 0 None 7 modEnvToPitch cent fs 0 -12000 -10 oct 12000 10 oct 0 None 8 initialFilterFc cent 8.176 Hz 1500 20 Hz 13500 20 kHz 13500 Open 9 initialFilterQ cB 0 0 None 960 96 dB 0 None 10 modLfoToFilterFc cent fs 0 -12000 -10 oct 12000 10 oct 0 None 11 modEnvToFilterFc cent fs 0 -12000 -10 oct 12000 10 oct 0 None 12 endAddrsCoarseOffset+ 32k smpls 0 * * 0 None 0 None 13 modLfoToVolume cbB fs 0 -960 -96 dB 960 96 dB 0 None 15 chorusEffectsSend 0.1% 0 0 None 1000 100% 0 None 16 reverbEffectsSend 0.1% 0 0 None 1000 100% 0 None 17 pan 0.1% Center -500 Left +500 Right 0 Center 21 delayModLFO timecent 1 sec -12000 1 msec 5000 20 sec -12000 <1 msec 22 freqModLFO cent 8.176 Hz -16000 1 mHz 4500 100 Hz 0 8.176 Hz 23 delayVibLFO timecent 1 sec -12000 1 msec 5000 20 sec -12000 <1 msec 24 freqVibLFO cent 8.176 Hz -16000 1 mHz 4500 100 Hz 0 8.176 Hz 25 delayModEnv timecent 1 sec -12000 1 msec 5000 20 sec -12000 <1 msec 26 attackModEnv timecent 1 sec -12000 1 msec 8000 100sec -12000 <1 msec 27 holdModEnv timecent 1 sec -12000 1 msec 5000 20 sec -12000 <1 msec 28 decayModEnv timecent 1 sec -12000 1 msec 8000 100sec -12000 <1 msec 29 sustainModEnv -0.1% attk peak 0 100% 1000 0% 0 attk pk 30 releaseModEnv timecent 1 sec -12000 1 msec 8000 100sec -12000 <1 msec 31 keynumToModEnvHold tcent/key 0 -1200 -oct/ky 1200 oct/ky 0 None 32 keynumToModEnvDecay tcent/key 0 -1200 -oct/ky 1200 oct/ky 0 None 33 delayVolEnv timecent 1 sec -12000 1 msec 5000 20 sec -12000 <1 msec 34 attackVolEnv timecent 1 sec -12000 1 msec 8000 100sec -12000 <1 msec 35 holdVolEnv timecent 1 sec -12000 1 msec 5000 20 sec -12000 <1 msec 36 decayVolEnv timecent 1 sec -12000 1 msec 8000 100sec -12000 <1 msec 37 sustainVolEnv cB attn attk peak 0 0 dB 1440 144dB 0 attk pk 38 releaseVolEnv timecent 1 sec -12000 1 msec 8000 100sec -12000 <1 msec 39 keynumToVolEnvHold tcent/key 0 -1200 -oct/ky 1200 oct/ky 0 None 40 keynumToVolEnvDecay tcent/key 0 -1200 -oct/ky 1200 oct/ky 0 None 43 keyRange MIDI ky# key# 0 0 lo key 127 hi key 0-127 full kbd 44 velRange MIDI vel 0 0 min vel 127 mx vel 0-127 all vels 45 startloopAddrsCoarseOffset+ smpls 0 * * * * 0 None 46 keynum+ MIDI ky# key# 0 0 lo key 127 hi key -1 None 47 velocity+ MIDI vel 0 1 min vel 127 mx vel -1 None 48 initialAttenuation cB 0 0 0 dB 1440 144dB 0 None 50 endloopAddrsCoarseOffset+ smpls 0 * * * * 0 None 51 coarseTune semitone 0 -120 -10 oct 120 10 oct 0 None 52 fineTune cent 0 -99 -99cent 99 99cent 0 None 54 sampleModes+ Bit Flags Flags ** ** ** ** 0 No Loop 56 scaleTuning cent/key 0 0 none 1200 oct/ky 100 semitone 57 exclusiveClass+ arbitrary# 0 1 -- 127 -- 0 None 58 overridingRootKey+ MIDI ky# key# 0 0 lo key 127 hi key -1 None * Range depends on values of start, loop, and end points in sample header. ** Range has discrete values based on bit flags + This generator is only valid at the instrument split level. 8.2 Source Enumerators In SoundFont 2.00, no modulators have yet been defined, hence no Source Enumerators have yet been established. Future Source Enumerators will include MIDI key number, MIDI key velocity, MIDI pitch bend, MIDI channel pressure, and various MIDI continuous controllers. 8.3 Transform Enumerators In SoundFont 2.00, no modulators have yet been defined, hence no Transform Enumerators have yet been established. Future Transform Enumerators will include a linear, or ÒnullÓ transform, as enumerator 0, and other transforms as required to implement the current default modulators. 8.4 Default Modulators In SoundFont 2.00, although no modulators have yet been defined, certain default modulation paths are expected to be implemented as standard. In a future revision to SoundFont 2, these default modulators will be implemented within the modulator structure, allowing variations from the defaults. The current default modulators are described below: 8.4.1 MIDI Key Velocity to Initial Attenuation The MIDI key number is passed through a ÒMIDI velocity to dBÓ transform to change it to the standard dB volume attentuation curve for MIDI velocity. A mathematical model for this transform will be included in a future version of this specification. The transformed result is then added to the initial attenuation generator. 8.4.2 MIDI Key Velocity to Filter Cutoff The MIDI key number is passed through a ÒMIDI velocity to FcÓ transform. A mathematical model for this transform will be included in a future version of this specification. The resulting value is added to the initial filter cutoff. This default modulator does not occur unless the volume envelope attack time is less than 7 msec. 8.4.3 MIDI Channel Pressure to Vibrato LFO to Pitch The MIDI Channel Pressure value value is divided by 128 to give a scaled value from zero to 127/128. This value is then multiplied by 18 and added to the Vibrato to LFO Pitch generator to give an increase in vibrato depth of 18 cents full scale. 8.4.4 MIDI Continuous Controller 1 to Vibrato LFO to Pitch The MIDI Continuous Controller 1 value is divided by 128 to give a scaled value from zero to 127/128. This value is then multiplied by 18 and added to the Vibrato to LFO Pitch generator to give an increase in vibrato depth of 18 cents full scale. 8.4.5 MIDI Continuous Controller 7 to Initial Attenuation The MIDI Continuous Controller 7 value is passed through a ÒMIDI velocity to dBÓ transform to change it to the standard dB volume attentuation curve for MIDI velocity. A mathematical model for this transform will be included in a future version of this specification. The transformed result is then added to the initial attenuation generator. 8.4.6 MIDI Continuous Controller 10 to Pan Position The MIDI Continuous Controller 10 value is added to -64, and the result divided by 64 to produce a scaled value ranging from -1 to +63/64. This value is then multiplied by 1000 and added to the pan generator to give a change in pan position from -100% to +100%. 8.4.7 MIDI Continuous Controller 11 to Initial Attentuation The MIDI Continuous Controller 11 value is passed through a ÒMIDI expressionÓ transform to change it to the standard dB volume attentuation curve for MIDI CC11. A mathematical model for this transform will be included in a future version of this specification. The transfromed result is then added to the initial attenuation generator. 8.4.8 MIDI Continuous Controller 91 to Reverb Effects Send The MIDI Continuous Controller 91 value is divided by 128 to give a scaled value from zero to 127/128. This value is then multiplied by 1000 and added to the reverbEffectsSend generator to give a change in reverb send from zero to +100%. 8.4.9 MIDI Continuous Controller 93 to Chorus Effects Send The MIDI Continuous Controller 93 value is divided by 128 to give a scaled value from zero to 127/128. This value is then multiplied by 1000 and added to the chorusEffectsSend generator to give a change in chrous send from zero to +100%. 8.5 Precedence and Absolute and Relative values. Most SoundFont generators are available at both the Instrument and Preset Levels, as well as having a default value. Generators at the Instrument Level are considered ÒabsoluteÓ and determine an actual physical value for the associated synthesis parameter, which is used instead of the default. For example, an value of 1200 for the attackVolEnv generator would produce an absolute time of 1200 timecents or 2 seconds of attack time for the volume envelope, instead of the default value of -12000 timecents or 1 msec. Generators at the Preset Level are instead considered ÒrelativeÓ and additive to all the default or instrument level generators within the Preset Layer. For example, a value of 2400 timecents for the attackVolEnv generator in a preset layer containing an instrument with two splits, one with the default attackVelEnv and one with an absolute attackVolEnv generator value of 1200 timecents would cause the default split to actually have a value of -9600 timecents or 4 msec, and the other to have a value of 3600 timecents or 8 seconds attack time. There are some generators which are not available at the Preset Level. These are: # Name 0 startAddrsOffset 1 endAddrsOffset 2 startloopAddrsOffset 3 endloopAddrsOffset 4 startAddrsCoarseOffset 12 endAddrsCoarseOffset 45 startloopAddrsCoarseOffset 46 keynum 47 velocity 50 endloopAddrsCoarseOffset 54 sampleModes 57 exclusiveClass 58 overridingRootKey If these generators are encountered in the Preset Level, they should be ignored. 9 Parameters and Synthesis Model The SoundFont 2 standard has been established with the intent of providing support for an expanding base of wavetable based synthesis models. The model supported by the SoundFont 2 specification originates with the EMU8000 wavetable synthesizer chip. The description below of the underlying synthesis model and the associated parameters are provided to allow mapping of this synthesis model onto other hardware platforms. 9.1 Synthesis Model The SoundFont 2 specification Synthesis Model comprises a wavetable oscillator, a dynamic lowpass filter, an enveloping amplifier, and programmable sends to pan, reverb, and chorus effects units. An underlying modulation engine comprises two low frequency oscillators (LFOs) and two envelope generators with appropriate routing amplifiers. 9.1.1 Wavetable Oscillator The SoundFont 2 specificationWavetable oscillator model is capable of playing back a sample at an arbitrary sampling rate with an arbitrary pitch shift. In practice, the upward pitch shift (downward sample rate conversion) will be limited to a maximum value, typically at least two octaves. The pitch is described in terms of an initial pitch shift which is based on the sampleÕs sampling rate, the root key at which the sample should be unshifted on the keyboard, the coarse, fine, and correction tunings, the effective MIDI key number, and the keyboard scale factor. All modulations in pitch are in octaves, semitones, and cents. 9.1.2 Sample Looping The wavetable oscillator is playing a digital sample which is described in terms of a start point, end point, and two points describing a loop. The sound can be flagged as unlooped, in which case the loop points are ignored. If the sound is looped, in can be played in two ways. If it is flagged as Òloop during releaseÓ, the sound is played from the start point through the loop, and loops until the note becomes inaudible. If not, the sound is played from the start point through the loop, and loops until the key is released. At this point, the next time the loop end point is reached, the sound continues through the loop end point and plays until the end point is reached, at which time audio is terminated. 9.1.3 Lowpass Filter The synthesis model contains a resonant lowpass filter, which is characterized by a dynamic cutoff frequency and a fixed resonance (Q). Because there is tremendous variation within the industry as to filter implementations, this filter is idealized rather than being specified as a particular realization. The filter is idealized at zero resonance as having a flat passband to the cutoff frequency, then a rolloff at 6dB per octave above that frequency. The resonance, when non-zero, comprises a peak at the cutoff frequency, superimposed on the above response. The resonance is measured as a dB ratio of the resonant peak to the DC gain. The DC gain at any resonance is half of the resonance value below the DC gain at zero resonance; hence the peak height is half the resonance value above DC gain at zero resonance. All modulations in cutoff frequency are in octaves, semitones, and cents. Figure 1 - Ideal Filter Response 9.1.4 Final Gain Amplifier The final gain amplifier is a multiplier on the filter output, which is controlled by an initial gain in dB. This is added to the volume envelope. Additional modulation can also be added. The gain is always specified in dB. 9.1.5 Effects Sends The output of the final gain amplifier can be routed into the effects unit. This unit causes the sound to be located (panned) in the stereo field, and a degree of reverberation and chorus to be added. The pan is specified in terms of percentage left and right, which also could be considered as an azimuth angle. The reverb and chorus sends are specified as a percentage of the signal amplitude to be sent to these units, from 0% to 100%. 9.1.6 Low Frequency Oscillators The synthesis model provides for two low frequency oscillators (LFOs) for modulating pitch, filter cutoff, and amplitude. The ÒvibratoÓ LFO is only capable of modulating pitch. The ÒmodulationÓ LFO can modulate any of the three parameters. An LFO is defined as having a delay period during which its value remains zero, followed by a triangular waveshape ramping linearly to positive one, then downward to negative 1, then upward again to positive one, etc. Each parameter can be modulated to a varying degree, either positively or negatively, by the associated LFO. Modulations of pitch and cutoff are in octaves, semitones, and cents, while modulations of amplitude are in dB. The degree of modulation is specified in cents or dB for the full scale positive LFO excursion. 9.1.7 Envelope Generators The synthesis model provides for two envelope generators. The volume envelope generator controls the final gain amplifier and hence determines the volume contour of the sound. The modulation envelope can control pitch and/or filter cutoff. An envelope generates a control signal in six phases. When key-on occurs, a delay period begins during which the envelope value is zero. The envelope then rises in a convex curve to a value of one during the attack phase. When a value of one is reached, the envelope enters a hold phase during which it remains at one. When the hold phase ends, the envelope enters a decay phase during which its value decreases linearly to a sustain level. When the sustain level is reached, the envelope enters sustain phase, during which the envelope stays at the sustain level. Whenever a key-off occurs, the envelope immediately enters a release phase during which the value linearly ramps from the current value to zero. When zero is reached, the envelope value remains at zero. Modulation of pitch and filter cutoff are in octaves, semitones, and cents. These parameters can be mdoulated to varying degree, either positively or negatively, by the modulation envelope. The degree of modulation is specified in cents for the full scale attack peak. The volume envelope operates in dB, with the attack peak providing a full scale output, appropriately scaled by the initial volume. The zero value, however, is actually zero gain. The implementation in the EMU8000 provides for 96 dB of amplitude control. When 96 dB of attenuation is reached in the final gain amplifier, an abrupt jump to zero gain (infinite dB of attenuation) occurs. In a 16 bit system, this jump is inaudible. 9.1.8 Modulation Interconnection Summary The following diagram shows the interconnections expressed in the SoundFont 2 specification synthesis model: Figure 2 - Modulation Structure 9.2 MIDI Functions The response to certain MIDI commands is defined within the MIDI spec, and is not considered to be part of the SoundFont 2 specification. For completeness, the expected response is given here. MIDI key number to Pitch - Relative to the Root Key as determined by the sample header byOriginalPitch and the Instrument Split overridingRootKey, keynumber varies the pitch by one cent times the scaleTune generator per keynumber. MIDI Pitch Bend - Changes the pitch of all notes in the MIDI channel as specified by the sensitivity set by the MIDI synthesizer mode or by using MIDI Registered Parameters. MIDI CC64 Sustain - ACTIVE when greater than or equal to 64. When the sustain function is active, all notes in the key-on state remain in the key-on state regardless of whether a key-off command for the note arrives. The key-off commands are stored, and when sustain becomes inactive, all stored key-off commands are executed. MIDI CC66 Soft - ACTIVE when greater than or equal to 64. When active, all new key-ons are modulated in such a way to make the note sound Òsoft.Ó This typically affects initial attentuation and filter cutoff is a pre-defined manner. MIDI CC67 Sostenuto - ACTIVE when greater than or equal to 64. When sostenuto becomes active, all notes currently in the key-on state remain in the key-on state until the sostenuto becomes inactive. All other notes behave normally. Notes maintained by sostenuto in key-on state remain in key-on state even if sustain is switched on and off. 9.3 Parameter Units The units with which SoundFont generators are described are all well defined. The strict definitions appear below: ABSOLUTE SAMPLE DATA POINTS - A numeric index of 16 bit sample data point words as stored in ROM or supplied in the smpl-ck, indexing the first sample data point word of memory or the chunk as zero. RELATIVE SAMPLE DATA POINTS - A count of 16 bit sample data point words based on an absolute sample data point reference. A negative value implies a relative count toward the beginning of the data. ABSOLUTE SEMITONES - A absolute logarithmic measure of frequency based on a reference of MIDI key numbers. A semitone is 1/12 of an octave, and value 69 is 440 Hz (A-440). Negative values and values above 127 are allowed. RELATIVE SEMITONES - A relative logarithmic measure of frequency ratio based on units of 1/12 of an octave, which is the twelfth root of two, approximately 1.059463094. ABSOLUTE CENTS - An absolute logarithmic measure of frequency based on a reference of MIDI key number scaled by 100. A cent is 1/1200 of an octave, and value 6900 is 440 Hz (A-440). Negative values and values above 12700 are allowed. RELATIVE CENTS - A relative logarithmic measure of frequency ratio based on units of 1/1200 of an octave, which is the twelve hundredth root of two, approximately 1.000577790. ABSOLUTE CENTIBELS - An absolute measure of the attenuation of a signal, based on a reference of zero being no attentuation. A centibel is a tenth of a decibel, or a ratio in signal amplitude of the two hundredth root of 10, approximately 1.011579454. RELATIVE CENTIBELS - A relative measure of the attenuation of a signal. A centibel is a tenth of a decibel, or a ratio in signal amplitude of the two hundredth root of 10, approximately 1.011579454. ABSOLUTE TIMECENTS - An absolute measure of time, based on a reference of zero being one second. A timecent represents a ratio in time of the twelve hundredth root of two, approximately 1.011579454. RELATIVE TIMECENTS - A relative measure of time ratio, based on a unit size of the twelve hundredth root of two, approximately 1.011579454. ABSOLUTE PERCENT - An absolute measure of gain, based on a reference of unity. In SoundFont 2, absolute percent is measured in 0.1% units, so a value of zero is 0% and a value of 1000 is 100%. RELATIVE PERCENT - A relative measure of gain difference. In SoundFont 2, relative percent is measured in 0.1% units. When the gain goes below zero, zero is assumed; when the gain exceeds 100%, 100% is used. 9.4 On Implementation Accuracy While the SoundFont 2 standard is well defined, it must be recognized that there are a large variety of practices and features within the wavetable music synthesis industry that are not conducive to exact implementation of the specification as defined. Some examples of impediments include the order of interpolation of sample data points, the exact shape and number of segments of envelopes, the filter implementation, and the details of the implementation of loops. Additionally, all real implementations are likely to have less accuracy than the SoundFont 2 standard itself. The units for the standard have been chosen to exceed the accuracy required for high fidelity applications. It should be recognized that in rendering a SoundFont 2 compatible file, a best practical reproduction is all that is expected. As such, implementers of SoundFont 2 compatible rendering engines will have to determine based on their own perceptual criteria the degree to which their implementation meets the standard. Approximations may take a variety of forms. In many cases, the resolution of the rendering engine will be less than that of the corresponding SoundFont unit. Also, it will frequently be the case that a line segment approximation will be made to a continuous curve. In the case of filters, the order of the filter may vary from the SoundFont 2 standard, and an optimum audible equivalent will have to be heuristically constructed. All such problems are left to the ingenuity of the implementers. 10 Error Handling 10.1 Structural Errors Structural Errors are errors which are determined from the implicit redundancy of the SoundFont RIFF file structure, and indicate that the structure is not intact. Examples are incorrect lengths for the chunks or subchunks, pointers out of valid range, or missing required chunks or subchunks for which no error correction procedure exists. In all cases, files should be checked for structural errors at load time, and if any are found the files should be rejected. Separate tools or options can be used to ÒrepairÓ structurally defective files, but these tools should validate that the reconstructed file is not only a valid SoundFont compatible bank but also complies with the intended timbral results in all cases. 10.2 Unknown Chunks In parsing the RIFF structure, unknown but well formed chunks or subchunks may be encountered. Unknown chunks within the INFO-list chunk should simply be ignored. Other unknown chunks or subchunks are illegal and should be treated as structural errors. 10.3 Unknown Enumerators Unknown enumerators may be encountered in Generators, Modulator Sources, or Transforms. This is to be expected if the ifil field exceeds the specification to which the application was written. Even if unexpected, unknown enumerators should simply cause the associated Generator or Modulator to be ignored. 10.4 Illegal Parameter Values Some SoundFont parameters are defined for only a limited range of the possible values which can be expressed in their field. If the value of the field is not in the defined range, the parameter has an illegal value. Illegal values for may be detected either at load or at run time. If detected at load time, the file may optionally be rejected as structurally unsound. If detected at run time, the default value for the parameter should be used if the parameter is required, or the entire Generator or Modulator ignored if it is optional. Certain parameters may have more specific procedures for illegal values as expressed elsewhere in this specification. 10.5 Out-of-range Values SoundFont parameters have a specified minimum and useful range the span the perceptually relevant values for the associated sonic property. When the parameter value is exceeds this useful range, the parameter is said to have an out of range value. Out of range values can result from two distinct causes. An out of range value can be actually present as a SoundFont generator value, or the out of range value can be the result of the summation of instrument and preset values. Out of range values should be handled by substituting the nearest perceptually relevant or realizable value. SoundFont compatible banks should not be created with out of range values in the instrument generators. While it is acceptable practice to create SoundFont banks which produce out of range values as a result of summation, it is undesirable and should be avoided where practical. 10.6 Missing Required Parameter or Terminator Certain parameters and terminators are required by the SoundFont specification. If these are missing, the file is technically not within specification. If such a problem is detected at load time, the file may optionally be rejected as structurally unsound. If detected at run time, the instrument or split for which the required parameter is missing should simply be ignored. If this causes no sound, the coresponding key-on event is ignored. 10.7 Illegal enumerator Certain enumerators are illegal in certain contexts. For example, key and velocity ranges must be the first generators in a layer or split, instruments are not allowed in splits, and sampleIDs are not allowed in layers. If such a problem is detected at load time, the file may optionally be rejected as structurally unsound. If detected at run time, the enumerator should simply be ignored. 11 Silicon SoundFonts 11.1 Silicon SoundFont Overview A ÒSilicon SoundFont BankÓ is an implementation of a SoundFont compatible bank realized in non- volatile memory with slight format additions. On initialization of a system using a Silicon SoundFont, the host processor navigates the Silicon SoundFont ROM format in sample memory space, determines the number of SoundFont Banks installed, and, when appropriate, reads the articulation data of the SoundFont files out of the Preset Data Chunks into its local RAM. The sample headers in the Silicon SoundFont point to the sample address offsets relative to the start of the Sample Chunk in the SoundFont compatile bank. The loader adds the appropriate offset to the sample addresses as part of its data management. Then, the system operates like any other SoundFont compatible system. The format of a Silicon SoundFont file intended to be burned into non-volatile memory is a hybrid between a standard ROM header and a modification of the standard SoundFont compatible bank file format. The ROM header contains data used for diagnostic tests, a ROM name, a size, and checksum information, and a sine wave sample to test audio outputs of a circuit. This is the first block of data found in the SoundFont ROM (address 0). The structure of the data contained in the ROM header is shown below. Because sample mamory space is word oriented, the endian nature of the resulting word reads is processor independent. However, the ogranization of bytes within a word, or words within a doubleword may vary on both the way the data has been encoded in the ROM and the endian nature of the processor. To handle all eventualities, it is recommended that the initialization software recognize and adapt for endian variations. 11.2 Silicon SoundFont ROM Header Format typedef struct romHdrType{ DWORD romRsrc; // unused DWORD romByteSize; // ROM size in bytes CHAR interleaveIndex; // for use in case of interleaved ROMs CHAR revision[3]; // for revision control CHAR id[4]; // matched with the IROM chunk in SF file format SHORT checksum; // to check ROM integrity SHORT checksum2sComplement; // for updating checksum variable w/o changing file // checksum value CHAR bankFormat; // unused CHAR product[16]; // product name (either system or SoundFont) BYTE sampleCompType; // indicates type of sample precompensation used CHAR filler1[2]; // future use CHAR style[16]; // sound library style CHAR copyright[80]; // copyright notice DWORD sampleStart; // beginning byte address of the SoundFont bank DWORD sineWaveStart; // beginning byte address of the sine wave sample DWORD filler2[124]; // future use SHORT sineWave[SINEWAVESIZE]; // sine wave sample data } romHdr; 12 Glossary absolute - Describes a parameter which gives a definitive real-world value. Contrast to relative. additive - Describes a parameter which is to be numerically added to another parameter. articulation - The process of modulation of amplitude, pitch, and timbre to produce an expressive musical note. artifact - A (typically undesirable) sonic event which is recognizable as not being present in the original sound. attack - That phase of an envelope or sound during which the amplitude increases from zero to a peak value. attenuation - A decrease in volume or amplitude of a signal. AWE32 - The original Creative Technology Sound Blaster product which contained an EMU8000 wavetable synthesizer and supported the SoundFont standard. bag - A SoundFont data structure element containing a list of layers (preset bag) or splits (instrument bag). balance - A form of stereo volume control in which both left and right channels are at maximum when the control is centered, and which attenuates only the opposite channel when taken to either extreme. bank - A collection of presets. See also MIDI bank. bidirectional compatibility - Simultaneous upward and downward compatibility. This refers to the fact that a properly designed SoundFont compatible program can appropriately handle files written to either a lower or higher revision of the specification. big endian - Refers to the organization in memory of bytes within a word such that the most significant byte occurs at the lowest address. Contrast Òlittle endian.Ó byte - A data structure element of eight bits without definition of meaning to those bits. BYTE - A data structure element of eight bits which contains an unsigned value from 0 to 255. case-sensitive - Indicates that an ASCII character or string treats alphabetic characters of upper or lower case as identical. Contrast Òcase-sensitive.Ó case-sensitive - Indicates that an ASCII character or string treats alphabetic characters of upper or lower case as distinct. Contrast Òcase-insensitive.Ó cent - A unit of pitch ratio corresponding to the twelve hundredth root of two, or one hundredth of a semitone, approximately 1.000577790. centibel - A unit of amplitude ratio corresponding to the two hundredth root of ten, or one tenth of a decibel, approximately 1.011579454. CHAR - A data structure of eight bits which contains a signed value from -128 to +127. chorus - An effects processing algorithm which involves cyclically shifting the pitch of a signal and remixing it with itself to produce a time varying comb filter, giving a perception of motion and fullness to the resulting sound. chunk - The top level division of a RIFF file. convex - A curve which is bowed in such a way that it is steeper on its lower portion. cutoff frequency - The frequency of a filter function at which the attenuation reaches a specified value. data points - The individual values comprising a sample. Sometimes also called sample points. Contrast Òsample.Ó decay - The portion of an envelope or sound during which the amplitude declines from a peak to steady state value. decibel - A unit of amplitude ratio corresponding to the twentieth root of ten, approximately 1.122018454. delay - The portion of an envelope or LFO function which elapses from a key-on event until the amplitude becomes non-zero. destination - The generator to which a modulator is applied. DC gain - The degree of amplification or attentuation a system presents to a static or zero frequency signal. digital audio - Audio represented as a sequence of quantized values spaced evenly over time. The values are called Òsample data points.Ó doubleword - A data structure element of 32 bits without definition of meaning to those bits. downloadable - Said of samples which are loaded from a file into RAM, in contrast to samples which are maintained in ROM. dry - Refers to audio which has not received any effects processing such as reverb or chorus. DWORD - A data structure of 32 bits which contains an unsigned value from zero to 4,294,967,295. EMU8000 - A wavetable synthesizer chip designed by E-mu Systems for use in Creative Technology products. envelope - A time varying signal which typically controls the pitch, volume, and/or filter cutoff frequency of a note, and comprises multiple phases including attack, decay, sustain, and release. enumerated - Said of a data element whose symbols correspond to particular assigned functions. extensible - Said of a format whose feature set can be expanded without impact on existing function. flat - A. Said of a tone that is lower in pitch than another reference tone. B. Said of a frequency response that does not deviate significantly from a single fixed gain over the audio range. generator - In the SoundFont standard, a parameter which directly affects sound reproduction. Contrast with Òmodulator.Ó global - Refers to parameters which affect all associated structures. See Òglobal layerÓ and Òglobal split.Ó global layer - A layer whose generators and modulators affect all other layers within the preset. global split - A split whose generators and modulators affect all other splits within the instrument. header - A data structure element which describes several aspects of a SoundFont element. hydra - A. A nine-headed mythical beast. B. The nine ÒpdtaÓ subchunks which make up the SoundFont articulation data. instrument - In the SoundFont standard, a collection of splits which represents the sound of a single musical instrument or sound effect set. instrument split - A sample and associated articulation data defined to play over certain key numbers and velocities. Also simply called a split. interpolator - A circuit or algorithm which computes intermediate points between existing sample data points. This is of particular use in the pitch shifting operation of a wavetable synthesizer, in which these intermediate points represent the output samples of the waveform at the desired pitch transposition. key number - See MIDI key number. layer - A subset of a preset containing generators, modulators, and an instrument. Also termed Òpreset layer.Ó level - In the SoundFont structure, this refers either to the preset and layers (the preset level) or the instrument and splits (the instrument level). LFO - Acronym for Low Frequency Oscillator. A slow periodic modulation source. linear coding - The most common method of encoding amplitudes in digital audio in which each step is of equal size. little endian - A method of ordering bytes within larger words in memory in which the least significant byte is at the lowest address. Contrast Òbig endian.Ó loop - In wavetable synthesis, a portion of a sample which is repeated many times to increase the duration of the resulting sound. loop points - The sample data points at which a loop begins and ends. lowpass - Said of a filter which attenuates high frequencies but does not attenuate low frequencies. monotonic - Continuously increasing or decreasing. Said of a sequence which never reverses direction. MIDI - Acronym for Musical Instrument Digital Interface. The standard protocol for sending performance information to a musical synthesizer. MIDI bank - A group of up to 128 presets selected by a MIDI Òchange bankÓ command. MIDI continuous controller - A construct in the MIDI protocol. MIDI key number - A construct in the MIDI protocol which accompanies a MIDI key-on or key-off command and specifies the key of the musical instrument keyboard to which the command refers. MIDI pitch bend - A special MIDI construct akin to the MIDI continuous controllers which controls the realtime value of the pitch of all notes played in a MIDI channel. MIDI preset - A ÒpresetÓ selected to be active in a particular MIDI channel by a MIDI Òchange presetÓ command. MIDI velocity - A construct in the MIDI protocol which accompanies a MIDI key-on or key-off command and specifies the speed with which the key was pressed or released. modulator - In the SoundFont standard, a set of parameters which affect a particular generator. Contrast with Ògenerator.Ó mono - Short for Òmonophonic.Ó Indicates a sound comprising only one channel or waveform. Contrast with Òstereo.Ó octave - A factor of two in ratio, typically applied to pitch or frequency. orphan - Said of a data structure which under normal circumstances is referenced by a higher level, but in this particular instance is no longer linked. Specifically, it is an instrument which is not referenced by any preset layer, or a sample which is not referenced by any instrument split. oscillator - In wavetable synthesis, the wavetable interpolator is considered an oscillator. pan - Short for Òpanorama.Ó This is the control of the apparent azimuth of a sound source over 180 degrees from left to right. It is generally implemented by varying the volume at the left and right speakers. pitch - The perceived value of frequency. Generally can be used interchangably with frequency. pitch shift - A change in pitch. Wavetable synthesis relies on interpolators to cause pitch shift in a sample to produce the notes of the scale. pole - A mathematical term used in filter transform analysis. Traditionally in synthesis, a pole is equated with a rolloff of 6dB per octave, and the rolloff of a filter is specified in Òpoles.Ó Preditor - E-mu SystemsÕ proprietary SoundFont 2.00 compatible bank editing software. preset - A keyboard full of sound. Typically the collection of samples and articulation data associated with a particular MIDI preset number. preset layer - A subset of a preset containing generators, modulators, and an instrument. Also simply termed a Òlayer.Ó proximal - Closest to. Proximal sample data points are the data points closest in either direction to the named point. Q - A mathematical term used in filter transform analysis. Indicates the degree of resonance of the filter. In synthesis terminology, it is synonymous with resonance. RAM - Random Access Memory. Conventionally, this term implies read-write memory. Contrast ÒROM.Ó record - A single instrance of a data structure. relative - Describes a parameter which merely indicates an offset from an otherwise established value. Contrast to absolute. release - The portion of an envelope or sound during which the amplitude declines from a steady state to zero value or inaudibility. resonance - Describes the aspect of a filter in which particular frequencies are given significantly more gain than others. The resonance can be measured in dB above the DC gain. resonant frequency - The frequency at which resonance reaches its maximum. reverb - Short for reverberation. In synthesis, a synthetic signal processor which adds artificial spaciousness and ambience to a sound. RIFF - Acronym for Resource Interchange File Format. The recommended form for interchange files such as SoundFont compatible files within Microsoft operating systems. ROM - Acronym for Read Only Memory. A memory whose contents are fixed at manufacture, and hence cannot be written by the user. Contrast with RAM. sample - This term is often used both to indicate a Òsample data pointÓ and to indicate a collection of such points comprising a digital audio waveform. The latter meaning is exclusively used in this specification. sample rate - The frequency, in Hertz, at which sample data points are taken when recording a sample. semitone - A unit of pitch ratio corresponding to the twelfth root of two, or one twelfth of an octave, approximately 1.059463094. sharp - Said of a tone that is higher in pitch than another reference tone. SHORT - A data structure element of sixteen bits which contains a signed value from -32,768 to +32,767. soft - The pedal on a piano, so named because it causes the damper to be lowered in such a way as to soften the timbre and loudness of the notes. In MIDI, continuous controller #66 which behaves in a similar manner. sostenudo - The pedal on a piano which causes the dampers on all keys depressed to be held until the pedal is released. In MIDI, continuous controller #67 which behaves in a similar manner. sustain - The pedal on a piano which prevents all dampers on keys as they are depressed from being released. In MIDI, continuous controller #64 which behaves in a similar manner. SoundFont - A registered trademark of E-mu Systems, Inc, indicating files produced by E-mu which conform to the SoundFont standard file format. source - In a SoundFont modulator, the enumerator indicating the particular realtime value which the modulator will transform, scale, and add to the destination generator. split - A sample and associated articulation data defined to play over certain key numbers and velocities. Also called an instrument split. stereo - Literally indicating three dimensions. In this specification, the term is used to mean two channel stereophonic, indicating that the sound is composed of two independent audio channels, dubbed left and right. Constrast monophonic. subchunk - A division of a RIFF file below that of the chunk. synthesis engine - The hardware and software associated with the signal processing and modulation path for a particular synthesizer. synthesizer - A device capable of producing ideally arbitrary musical sound. terminator - A data structure element indicating the final element in a sequence. timecent - A unit of duration ratio corresponding to the twelve hundredth root of two, or one twelve hundredth of an octave, approximately 1.000577790. transform - In a SoundFont modulator, the enumerator indicating the particular transfer function through which the source will be passed prior to scaling and addition to the destination generator. tremolo - A periodic change in amplitude of a sound, typically produced by applying a low frequency oscillator to the final volume amplifier. triangular - A waveform which ramps upward to a positive limit, then downward at the opposite slope to the symmetrically negative limit periodically. unpitched - Said of a sound which is not characterized by a perceived frequency. This would be true of noise-like musical instruments and of many sound effects. velocity - In synthesis, the speed with which a keyboard key is depressed, typically proportionally to the impact delivered by the musician. See also MIDI velocity. vibrato - A periodic change in the pitch of a sound, typically produced by applying a low frequency oscillator to the oscillator pitch. volume - The loudness or amplitude of a sound, or the control of this parameter. wavetable - A music synthesis technique wherein musical sounds are recorded or computed mathematically and stored in a memory, then played back at a variable rate to produce the desired pitch. Additional timbral adjustments are often made to the sound thus produced using amplifiers, filters, and effect processing such as reverb and chorus. WORD - A data structure of 16 bits which contains an unsigned value from zero to 65,535. word - A data structure element of 16 bits without definition of meaning to those bits. SoundFon 2.00a Technical Specification - Page 1 - Printed 12/4/95 at 13:32