A SoundFont bank (or SoundFont compatible bank) contains both the digital
audio samples which have been captured from a sound source, and the
instructions to the wavetable synthesizer on how to articulate this sound
based on the musical or sonic context as expressed by MIDI (called
articulation data). A SoundFont bank also contains information about
the creation and intended use of the particular bank.
A SoundFont bank is stored in the industry standard RIFF format. There are 3
major portions or "chunks" contained within a SoundFont bank, one for
information, one for sample data, and one for articulation data.
The information chunk contains information about the bank. The Sample Data
chunk contains the digitized waveforms used in the SoundFont bank itself.
The articulation data chunk is a bit more involved.
The articulation data chunk uses two levels of articulation data which
sits atop a level of sample header data which describes sample data itself.
Each level of articulation data REFERENCES the level beneath it, thus allowing
for reuse of resources.
SF2 Articulation Data Hierarchy:
PRESET LEVEL
.
.
.
Preset m
Name
Variation Bank Index
Program Index
Layer n
Articulation data
Instrument i Reference
Layer n+1
Articulation data
Instrument h Reference
.
.
.
Preset m+1
Name
Variation Bank Index
Program Index
Layer o
Articulation data
Instrument p Reference
Layer o+1
Articulation data
Instrument a Reference
.
.
.
INSTRUMENT LEVEL
.
.
.
Instrument i
Name
Split j
Articulation Data
Sample s Reference
Split k
Articulation Data
Sample b Reference
.
.
.
Instrument i+1
Name
Split l
Articulation Data
Sample s Reference
Split l+1
Articulation Data
Sample t Reference
.
.
.
SAMPLE HEADER LEVEL
.
.
.
Sample s
Name
Location and loop points
Information
Sample s+1
Name
Location and loop points
Information
.
.
.
The lowest level of articulation data is called the "Sample Header" level.
It contains a list of data structures, each describing one particular waveform
contained within the SoundFont bank or a particular waveform contained within
a wavetable ROM. Each data structure has a unique name for each sample,
information regarding the location of the sample relative to the beginning
of the sample data contained within the bank (or the absolute location of the
sample in a particular wavetable ROM), sample loop points, original
sample rate of the digitized waveform, and other information about the samples.
The combination of the Sample Header data and the sample data for a single
sampled waveform in a SoundFont bank is the rough equivalent of a WAV file.
However, stereo SoundFont sounds are quite different than stereo WAV sounds.
See question below for details.
The next highest level of articulation data is called the "Instrument" level.
Instruments are subdivided into "Splits". A Split is the combination of a
key range and/or a velocity range, a reference to a particular sample header
within the SoundFont bank, and articulation data which is applied directly to
that sample.
By referencing samples, it is possible to have any split articulate any sample
in the bank.
IE:
INSTRUMENT LEVEL
Instrument 1:
Split 1
Key 0-127
Articulation Data
Sample q
.
.
.
Instrument y:
Split 1
Key 0-127
Articulation Data
Sample q
.
.
.
SAMPLE LEVEL
.
.
Sample q:
Loop points, etc
.
.
By allowing multiple splits within a single instrument, it is possible to have
different samples played back with varying keynumber and velocity, each with
independent articulation.
IE:
INSTRUMENT LEVEL
Instrument 1:
Split 1
Key 43-43
Articulation Data
Sample q
Split 2
Key 44-44
Vel 0-120
Other Articulation Data
Sample x
Split 3
Key 44-44
Vel 101-127
Other Articulation Data
Sample q
Split 4
Key 0-42
Other Articulation Data
Sample a
.
.
.
.
SAMPLE LEVEL
.
.
Sample a:
Loop points, etc
.
.
Sample q:
Loop points, etc
.
.
Sample x:
Loop points, etc
One can see that keys 0-42 plays sample a, and key 43 plays sample q. So key
range is variable. Also one can see that key number 44 plays sample x when
struck softly, sample q when struck hard, and BOTH samples when struck medium.
So velocity sensitive samples is supported.
These features can be used to form drumkits, finely tuned complex instruments
which do not sound "munchkin-like" after pitch shifting, or velocity dependent
sounds.
A combination of splits (which (typically) collectively span the range of
possible keynumbers and velocity values) is collected to form an "Instrument".
The highest level of articulation data is called the "Preset" level. Presets
are subdivided into "Layers". A Layer is the combination of a key range and/or
a velocity range, a reference to a particular instrument within the SoundFont
bank, and articulation data which is applied RELATIVE to corresponding
articulation data within that instrument.
By referencing instruments, it is possible to have any layer articulate any
instrument in the bank. By applying articulation data at the Preset level
RELATIVE to data at the Instrument level, it is possible to customize
instruments which are professionally designed and fine tuned without having
to destroy the original material.
IE:
PRESET LEVEL
Preset 1:
Layer 1
Key 0-127
Add i% Reverb
Instrument x
.
.
.
INSTRUMENT LEVEL
Instrument 1:
Split 1
.
.
.
Here, i% reverb is ADDED to WHATEVER the reverb is on WHATEVER samples are
used in the instrument!
By allowing multiple layers within a single Preset, it is possible to have
different instruments exist at different parts of the keyboard or in different
velocity ranges, each with independent articulation.
This can be used to make presets in which your left hand plays a bass guitar
and your right plays a piano:
PRESET LEVEL
Preset 1:
Layer 1
Key 0-63
Instrument "Bass Guitar"
Layer 2
Key 64-127
Instrument "Piano"
chorded versions of instruments:
PRESET LEVEL
Preset 1:
Layer 1
Key 0-127
Instrument "Piano"
Layer 2
Key 0-127
Add 4 semitones of pitch tuning
Instrument "Piano"
Layer 3
Key 0-127
Add 7 semitones of pitch tuning
Instrument "Piano"
velocity sensitive instruments:
PRESET LEVEL
Preset 1:
Layer 1
Key 0-127
Vel 0-100
Instrument "Piano"
Layer 2
Key 0-127
Vel 101-127
Instrument "Bass Guitar"
or any other custom variations of finely tuned instruments. All of that
without having the user worry about samples or loop points or copying of
large numbers of splits or other such complex issues. Expert users of Vienna
1.0 should be able to recognize that creating presets like the above
would be a nightmare with the SoundFont 1.0 format! (EsBeeKay users COULD
do the Piano/Bass Guitar, but not the chorded piano nor the add i% reverb)
A combination of layers (which (typically) collectively span the
range of possible keynumbers and velocity values) is collected to form a
"Preset".
Each Preset has a corresponding "variation bank" index and "program index"
which are directly analogous to MIDI "bank change" and "program change"
commands.
A SoundFont bank is viewed as the combination of a collection of Presets,
Instruments, and Samples, each self contained and only dependent on the level
beneath it.
IE:
A single sample header and specified sample data are viewed as an entity.
A single instrument and all of the samples it uses are viewed as an entity.
A single preset and all of the instruments it uses are viewed as an entity.
SoundFont banks do not contain "interleaved" samples, instead the sample
header data contains "sample links" in which individual samples "point" to other
samples which are their stereo counterpart (or which are linked in a chain of
samples.) These references specify pairs or groups of sampled waveforms which
are to be played in perfect pitch phase. Those samples are played in perfect
pitch phase IF both or all samples are used in the same INSTRUMENT. With this
approach, it is possible to use one sample in a stereo pair independently as
a mono sample or as one of a stereo pair of samples, or combinations of both
in different instruments within the same SoundFont bank.
Each sample header contains a field where it declares itself as a mono
or a master or a slave sample, and another field where it has a reference to
the other sample, or the next sample in the chain.
Stereo sounds are be produced when an INSTRUMENT contains two splits,
each of which point to the two stereo samples, and each of which are triggered
with the same key/velocity combination. These splits may have INDEPENDENT
articulation parameters settings (including pan position) EXCEPT those
involving pitch modulation.
IE:
Preset "Stereo Sound"
Layer 1
Instrument "Stereo Instrument"
Split 1
Articulation Data Parameters
Sample "Right Sample" (Master, points to "Left Sample")
Split 2
Articulation Data Parameters
Sample "Left Sample" (Slave, points to "Right Sample")
In the case of pitch, one split (called the MASTER split) controls the pitch
for its sample AND the sample in the OTHER SPLIT (or SLAVE split). So changing
pitch modulation parameters (such as LFO to Pitch or fine tuning) in a SLAVE
split does nothing, but changing such parameters in a MASTER split has that
modulation apply to BOTH samples!
There is no other "automatic" parameter settings in SoundFont stereo sounds.
Pan position may be wherever you like. (no automatic pan position of stereo
paired samples). Each sample may be played at the loudness they are both
individually capable of (no automatic attenuation of stereo paired sounds).
Each sample gets its own individual attenuation and filter settings! (no
automatic articulation data copying from master split to slave split in
stereo paired sounds). Thus, it is POSSIBLE to build a classic stereo sound
(pan hard left/right, half volume on each, same articulation on each) in a
SoundFont 2.0 bank but you are not CONFINED to those settings!
If a keynumber/velocity combination triggers only ONE split in an instrument
and that split contains ONE of a pair of stereo samples, then you hear ONLY
that sample in MONO. This way, you can use the samples individually or in
stereo if you like.
So if an INSTRUMENT only contains ONE sample in a stereo pair, that
is a MONO instrument.
IE:
Preset "Mono Sound"
Layer 1
Instrument "Mono Instrument"
Split 1
Articulation Data Parameters
Sample "Right Sample" (Master, points to "Left Sample")
Even though the "Right Sample" still points to the "Left Sample",the
"Left Sample" is never used in the "Mono Instrument". So that is a mono
sound with just the one sample!
If a PRESET contains two INSTRUMENTS, one of which holds only ONE
sample in a stereo pair, and the other of which holds only THE OTHER sample
in a stereo pair, a stereo sound does NOT result. That would have the effect
of FUNDAMENTALLY changing the nature of an instrument with nothing more than
the inclusion of another instrument. This defies the 'instrument as an entity'
property.
IE:
Preset "NOT A Stereo Sound"
Layer 1
Instrument "Mono Instrument"
Split 1
Articulation Data Parameters
Sample "Right Sample" (Master, points to "Left Sample")
Layer 2
Instrument "Other Mono Instrument"
Split 1
Articulation Data Parameters
Sample "Left Sample" (Slave, points to "Right Sample")
Even though the "Right Sample" still points to the "Left Sample", AND
the "Left Sample" IS used in another instrument, this is NOT a stereo
sound!
A single SoundFont bank may be made with all of the above examples:
PRESET LEVEL
Preset 1: "Stereo Sound"
Layer 1
Instrument 1
Preset 2: "Mono Sound"
Layer 1
Instrument 2
Preset 3: "NOT A Stereo Sound"
Layer 1
Instrument 2
Layer 2
Instrument 3
Preset 4: "Some other unrelated sound"
Layer 1
Instrument 4
INSTRUMENT LEVEL
Instrument 1: "Stereo Instrument"
Split 1
Articulation Data Parameters
Sample 1
Split 2
Articulation Data Parameters
Sample 2
Instrument 2: "Mono Instrument"
Split 1
Articulation Data Parameters
Sample 2
Instrument 3: "Other Mono Instrument"
Split 1
Articulation Data Parameters
Sample 2
Instrument 4: "Some other unrelated instrument"
Split 1
Articulation Data Parameters
Sample 3
SAMPLE LEVEL
Sample 1: "Right Sample" (Master, points to "Left Sample")
Loop points, etc
Master
Points to Sample 2
Sample 2: "Left Sample" (Slave, points to "Right Sample")
Loop points, etc
Slave
Points to Sample 1
Sample 3: "Some other sample"
Loop points, etc
Mono
Pointer is irrelevant
This example does not use PRESET LEVEL Articulation Data Parameters,
but you can do that too!
Finally, "stereo" sound in SoundFont 2.0 is merely a subset of general number
of pitch phase locked sample sounds. In the case of "stereo" sound, that
"general number" is 2. Thus it is possible to pitch phase lock as many voices
as you like, so long as the link information in the sample header is used to
form a CLOSED CHAIN of samples.
IE Sample 1 points to Sample 2 which points to Sample 3... which points
to Sample n which points back to Sample 1.
Creative Garners Top Awards in 1995
