Synthesis

Symusic can turn a Score into sound using SoundFont 2/3 data and the Prestosynth engine. This section explains the moving pieces so you can audition scores, render datasets, or embed procedural music into an application.

Components

  1. SoundFont – .sf2/.sf3 file that defines instrument samples and playback rules. Use BuiltInSF2/3 helpers to download curated fonts or provide your own path.

  2. Synthesizer – Python/nanobind wrapper around Prestosynth. Constructed with sf_path, sample_rate, and an optional quality flag.

  3. Score – Input data. Tracks contribute program and is_drum metadata so instruments map correctly.

  4. render() – Converts the score to seconds (if necessary), schedules each event, and returns a NumPy array shaped [channels, samples].

  5. dump_wav() – Convenience helper to write the array to disk.

Usage pattern

from symusic import Score, Synthesizer, BuiltInSF3, dump_wav

score = Score("piece.mid")
sf_path = BuiltInSF3.MuseScoreGeneral().path(download=True)
synth = Synthesizer(sf_path=sf_path, sample_rate=48000, quality=1)
audio = synth.render(score, stereo=True)
dump_wav("piece.wav", audio, sample_rate=48000)

  • Rendering always works in second units internally, so conversions are automatic.

  • Percussion tracks (is_drum=True) trigger the GM drum kit region inside the SoundFont.

  • quality toggles different interpolation / oversampling strategies inside Prestosynth.

When to reach for synthesis

  • Preview arrangements without leaving your notebook.

  • Generate audio datasets aligned with symbolic annotations.

  • Prototype differentiable or procedural music systems that need quick feedback loops.

See symusic.Synthesizer and symusic.dump_wav for all arguments and return types.