A specification-compliance test for SoundFont synthesizers.
Updated 12/31/25 by S. Christian Collins
HELP WANTED! If someone could send me audio of this test recorded on a Sound Blaster Live! with reverb and chorus enabled, I would be much obliged. You can upload your recording to this bug report.
The following tests will check to see if your SoundFont player is compatible with the SoundFont 2.01 and/or 2.04 specification. This test will not cover every possible implementation flaw, but should give a decent overview of a SoundFont player's compatibility with the specification.
Recordings of the test using several prominant SoundFont synthesizers are included in the recordings folder. Currently, the most standards-compliant SoundFont synth is FluidSynth, so I recommend using the FluidSynth 2.5.2, custom fx.flac recording as a reference when comparing any other test results. The Sound Blaster Audigy2 ZS, custom fx.flac recording is also a good reference as the best hardware implementation of the SoundFont 2.01 standard.
To run the test, load the accompanying SoundFont bank into your SoundFont synth as bank 0, and then play the MIDI file. Use Audigy2 ZS, custom fx.ogg in the recordings folder as a reference for how these tests should sound. Note that for some tests, FluidSynth provides a more ideal rendering, especially when it comes to the filter behavior.
This test uses all six envelope stages, each lasting exactly 1 second*. The stages are delay, attack, hold, decay, sustain, and release. The attack should be a convex curve, while the remaining envelope stages should all be linear. Note that when viewing these stages on a waveform view based on amplitude (e.g., default Audacity view), the attack will appear linear, and the other phases will appear concave. The test starts with a click. You should hear the note's volume change according to the following envelope shape (as viewed on a dB scale):
* – The decay speed is based on the time it would take to decay from 0 dB to -96 dB, but since the sustain level is -12 dB, the time spent actually decaying the sound is much shorter than 1 second.
This test uses all six envelope stages, each lasting exactly 1 second*. The stages are delay, attack, hold, decay, sustain, and release. The attack should be a convex curve, while the remaining envelope stages should all be linear. You should hear the note's pitch change according to the following envelope shape:
* – The decay speed is based on the time it would take to decay from 100% to 0% pitch modulation, but since the sustain level is 50%, the time spent actually decaying the sound is only half a second.
You should hear five tones at the same pitch, each one increasing in duration of the decay phase.
You should hear five tones at the same pitch, each one increasing in duration of the hold phase.
You should hear the volume oscillating at a moderate speed (4 Hz). The LFO depth value is set to 6 dB, which means the sample volume should oscillate between +6 dB and -6 dB. In a waveform view, the volume oscillation should appear as a triangle wave.
-
Test A: Velocity = 127. On Sound Blaster hardware with no attenuation to the sound at instrument or preset level, the sound cannot be further amplified. Therefore, the LFO will only cause a reduction in volume, but never a boost. In a waveform view, the volume peaks will appear to be "chopped off". See the notes for test 12 below. This volume limitation does not exist in FluidSynth or BASSMIDI.
-
Test B: Velocity = 90. Since there is attenuation to the sound due to the lower note velocity, the LFO should be able to both amplify and attenuate the sound regardless of SoundFont synth used.
You should hear the pitch oscillating at a moderate speed (4 Hz).
Which LFO is activated by the mod wheel (CC1)? According to the SoundFont 2.01 and 2.04 specs, the mod wheel should add 50 cents to the vibrato LFO (LFO2), resulting in a moderate-speed vibrato (4 Hz) in this test.
SoundFont 2.0 and earlier devices such as the Sound Blaster Live! or AWE32 had the mod wheel activating the modulation LFO (LFO1) instead, which will result in a very fast vibrato (8.176 Hz) in this test.
You should hear three tones, all at the same pitch.
You should hear a white noise wave filtered using a low-pass filter at each of the following values:
| Cutoff frequency (Hz) |
|---|
| 20 |
| 50 |
| 100 |
| 200 |
| 500 |
| 1,000 |
| 1,500 |
| 2,000 |
| 3,000 |
| 4,000 |
| 5,000 |
| 6,000 |
| 7,000 |
| 8,000 |
| 9,000 |
| 10,000 |
| 12,000 |
| 15,000 |
| 17,500 |
| 20,000 |
Note that Sound Blaster hardware has an unusual filter that is fully open at 8,000 Hz. FluidSynth implements a proper 2-pole lowpass filter. See my video here on this topic.
You should hear a white noise wave filtered using a low-pass filter at 4,000 Hz while increasing the filter resonance (Q) using the following values:
| Filter Q (dB) |
|---|
| 0 |
| 5 |
| 10 |
| 15 |
| 20 |
| 25 |
| 30 |
| 35 |
| 40 |
| 50 |
| 70 |
| 96 |
The Sound Blaster hardware filter resonance appears to be capped to a maximum of 20 dB.
Each 1 dB of attenuation set at the instrument or preset level should only attenuate the sound by 0.4 dB. This is a quirk of the Sound Blaster hardware that should be emulated for compatibility with existing SoundFonts. Recent versions of Polyphone translate this value in the editor so you can see the actual attenuation amount.
For this test, you will hear a series of tones getting progressively quieter, based on the table below:
| dB in SoundFont | actual dB attenuated |
|---|---|
| 0 | 0 |
| 5 | 2 |
| 10 | 4 |
| 15 | 6 |
| 20 | 8 |
| 25 | 10 |
| 30 | 12 |
When measured, the volume decrease between each tone should be exactly 2 dB.
Same as Test #11, but with preset attenuation set to -12 dB. All tones should play at the same volume.
On Sound Blaster hardware, a playing sample (or "voice") can either be at full volume (0 dB) or attenuated below that level. No modulator, LFO or attenuation parameter value can raise the sample playback volume above this 0 dB limit. It is possible to set negative attenuation values at the preset level. However, if a voice has not been attenuated by a modulator (e.g., velocity, CC7, CC11), envelope, or by the attenuation value at the instrument level, then it cannot be boosted. In other words, a negative attenuation value in the preset can only boost the voice volume up to 0 dB, but never beyond.
Measured at the following velocities: 127, 111, 95, 79, 63, 47, 31, 15.
- Test A: default (should be 96 dB concave. The difference between velocity 127 and 111 should be 2.34 dB)
- Test B: 144 dB concave
- Test C: 48 dB concave
- Test D: 96 dB linear
- Test E: modulator deleted from the instrument (all tones should be the same volume)
All tones should be the same volume, assuming test #13 is passed. Measured at the following velocities: 127, 111, 95, 79, 63, 47, 31, 15, with filter cutoff set to 4,000 Hz, and filter resonance at 10 dB.
The result of this test will vary depending on which spec version is implemented by the SoundFont synth:
-
2.01 spec: You should hear no filtering as the velocity decreases from 127 to 64, and then a sudden jump to -1200 cents filtering when the velocity reaches 63. From there, the filtering gradually increases as you approach velocity 0 to a total of -2400 cents. This sudden jump in filtering is caused by the default modulator's use of a negative unipolar switch as the secondary modulator source.
This version of the default velocity-to-filter-cutoff modulator was a mistake. From the description in section 8.4.2 of the SoundFont 2.01 spec, it is clear the developers were trying to recreate the behavior of the Sound Blaster AWE32/Live! filter whereby gradual filtering would occur from velocities 127 through 64, and then no further filtering would occur below velocity 64. However, this desired behavior is not possible with the described model, and is actually not possible at all using a negative unipolar primary modulator source. After the hardware implementation had been created (E-MU APS, Sound Blaster Audigy series), the developers realized their mistake and redesigned the default modulator for the SoundFont 2.04 spec.
-
2.04 spec: There shuld be moderate filtering (-2400 cent curve) as the velocity decreases from 127 to 0 with no sudden jump in the middle.
-
Earlier spec versions: You should hear no filtering. In earlier spec versions, filtering only occurs when the volume envelope attack is at 7 ms or higher (8 ms on the Sound Blaster Live!), and thus won't be triggered in this test.
Because of this inconsistent behavior and the different way each version of this modulator must be canceled/overridden by the SoundFont designer, many SoundFont synths including FluidSynth and BASSMIDI choose not to implement this default modulator at all. Regardless, I encourage all SoundFont designers to cancel both 2.01 and 2.04 versions of this modulator in their instruments to ensure consistent filter behavior when played in different synthesizers. Canceling these modulators is done at the instrument level (not preset), and in Polyphone looks like this (2.01 modulator on top, 2.04 modulator on bottom):
If you wish to use velocity-based filtering, then change the amount on the 2.04 modulator as desired (default = -2400 cents).
The result should be a smooth progression over a wide frequency range for the low pass filter, with noticeably more filtering at low velocities than the default velocity-to-FC modulator.
On the Sound Blaster Live!, a volume envelope attack of 8 ms or greater enables a velocity-to-FC modulator that reduces the filter cutoff from velocities 127 to 64, but then no further filter cutoff reductions happen through velocities 63 to 0. The same is true on AWE32/64 with volume envelope attack of 7 ms or greater. This test attempts to see if this behavior is emulated. If Test A does not result in a progressively filtered sound but Test C does, then this old behavior is emulated. If Test A sounds filtered, then ignore the results of this test.
Tests to see if the synth responds to the 2.01 spec method of canceling the default velocity-to-FC modulator. If so, all tones should have the same filtering. Ignore this test if the default modulator is not implemented (see Test A).
Tests to see if the synth responds to the 2.04 spec method of canceling the default velocity-to-FC modulator. If so, all tones should have the same filtering. Ignore this test if the default modulator is not implemented (see Test A).
You should hear a smooth opening and closing of the filter, which is being controlled by the mod wheel (CC1). If this isn't implemented, you will probably hear vibrato instead.
You will hear either "supported" or "not supported". Strangely, the Sound Blaster Live! does not support sample offset, even though the AWE32/64 does.
- Test A: Instrument-level reverb = 0%, 33.3%, 66.6% and 100%.
- Test B: MIDI CC 91 reverb = 0 (0%), 42 (33.3%), 85 (66.6%) and 127 (100%).
- Test C: Both instrument-level and MIDI CC91 reverb = 0%, 33.3%, 66.6% and 100%.
- Test A: Instrument-level chorus = 0%, 33.3%, 66.6% and 100%.
- Test B: MIDI CC 93 chorus = 0 (0%), 42 (33.3%), 85 (66.6%) and 127 (100%).
- Test C: Both instrument-level and MIDI CC 93 chorus = 0%, 33.3%, 66.6% and 100%.
This test must be performed by ear. If the sound is grainy, a low quality interpolation method (e.g., "linear") is probably being used.
- Test A: Default (+/- 2 semitones). You should hear middle C bend down a whole step and back, then up a whole step and back.
- Test B: Modulator removed. You should hear no pitch change.
- Test C: Inverted. You should hear middle C bend up a whole step and back, then down a whole step and back.
Very short noise = exclusive class is supported Long noise = exclusive class is not supported
Tests for correct panning behavior when using both mono and stereo samples. Each test will play nine tones progressing from left to right in the stereo field. A mono sample panned fully left (-50% internal to the SoundFont) should play only through the left speaker. A stereo sample panned fully left should have the left sample played only through the left speaker and the right channel played centered. This behavior should of course be mirrored when panning to the right (50% internally). The stereo sample for this test contains two sine waves an octave apart, the lower tone fully in the left channel and the higher tone fully in the right channel.
Please refer to this chart for the values of the nine tones played in the tests below:
| Test A/C (internal) | Test E (internal) | Test B/D/F (CC10) |
|---|---|---|
| -50% | -100% | 0 |
| -37.5% | -75% | 16 |
| -25% | -50% | 32 |
| -12.5% | -25% | 48 |
| 0% | 0% | 64 |
| 12.5% | 25% | 80 |
| 25% | 50% | 96 |
| 37.5% | 75% | 112 |
| 50% | 100% | 127 |
Mono sine wave using internal pan values -50% – 50%.
- -50% and 50% should result in fully left and right panned samples, respectively.
Mono sine wave using MIDI CC 10 (pan) values 0 – 127, which should correspond to the results in Test A above.
- Note that CC 10 value 127 corresponds to 49.22% (right) pan, not true 50%, though some synthesizers may compensate for this. Every other value should create a perfect sonic match with Test A above.
Stereo sine waves using internal pan values -50% – 50%.
- At -50% (left), you should hear the higher-pitched sine wave (the sample's right channel) playing centered—not fully panned left. Panned 50% (right), it is the lower-pitched sine wave (the sample's left channel) that should now be centered.
Stereo sine waves using MIDI CC 10 (pan) values 0 – 127, which should correspond to the results in Test C above.
- Note that CC 10 value 127 corresponds to 49.22% (right) pan, not true 50%, though some synthesizers may compensate for this. Every other value should create a perfect sonic match with Test C above.
Stereo sine waves using internal pan values -100% – 100%.
- At -100%, both channels of the stereo sample should be fully panned to the left. At 100%, everything is fully to the right.
By overriding the default CC10-to-pan modulator, it is possible to use MIDI CC 10 to replicate the results from Test E above.
- This test should sound identical to Test E above (with the aforementioned exception of CC value 127). However, if the SoundFont synth does not support overriding the default CC10-to-pan modulator, then it will sound identical to Test D instead.
Most of the sine wave tests peak at the same volume, and I have normalized this peak at -6 dB in all of the recordings I have provided with this project. Recordings provided by other users may or may not also be normalized in this way. A strange quirk, however, appears in the Sound Blaster recordings, where there are slight volume differences between many of these tests (e.g., the volume envelope test peaks at -5.251 dB, whereas the modulation envelope test peaks at -6 dB). I have scoured both the SoundFont and MIDI file for anything that could be the cause and have come up empty. My best guess as to why this happens is that there is some undefined behavior in the Sound Blaster volume envelope whereby certain parameters can cause an unexpected effect on the sample volume. More tests would need to be developed to find out for sure.