audiate

Render a symbolic music score to audio.

Give audiate a score in any of several standard formats — MIDI, MusicXML, ABC, Humdrum-kern, MEI — and get back an audio waveform. One call:

import audiate

audio = audiate.render("song.mid")  # -> AudioData
audio.write("song.wav")

The name is audiation — hearing music in your mind. audiate makes a written score actually audible.

Install

pip install audiate                 # core: MIDI in, pure-Python "sine" synth out
pip install "audiate[symbolic]"     # + MusicXML / ABC / kern / MEI (music21)
pip install "audiate[fluidsynth]"   # + SoundFont synthesis (pyFluidSynth)
pip install "audiate[all]"          # everything

Quick start

import audiate

# Start from any standard format — it's auto-detected
audio = audiate.render("lead_sheet.musicxml")
audio = audiate.render("tune.abc")
audio = audiate.render("bach.krn")

# ...or from objects you already have
import pretty_midi

audio = audiate.render(pretty_midi.PrettyMIDI("song.mid"))

from music21 import corpus

audio = audiate.render(corpus.parse("bach/bwv66.6"))

# The result is an AudioData
audio.array  # float32 numpy waveform
audio.sample_rate  # e.g. 44100
audio.duration_seconds
audio.to_wav_bytes()  # WAV bytes
audio.write("out.wav")  # write to disk

How it works

The universal pivot is MIDI. Every input format is normalized to a pretty_midi.PrettyMIDI (the ingest layer), then a pluggable engine synthesizes it to audio. Adding one ingester or one engine multiplies coverage across all the others.

score (MIDI / MusicXML / ABC / kern / MEI / …)
        │  ingest  (pretty_midi + music21)
        ▼
   pretty_midi.PrettyMIDI          ← the MIDI-in-the-middle pivot
        │  engine
        ▼
   AudioData (waveform + sample_rate)

Engines (progressive disclosure)

engine="auto" (the default) uses fluidsynth when a SoundFont is available, otherwise falls back to sine.

audio = audiate.render("song.mid", engine="fluidsynth", soundfont="MyPiano.sf2")
audio = audiate.render("score.musicxml", engine="musescore")  # highest fidelity

Register your own engine — it's an open registry:

from audiate import register_engine


@register_engine("my_synth")
def my_synth(pm, *, sample_rate=44100, **opts):
    ...
    return waveform, sample_rate  # (np.ndarray, int)

System dependencies

• FluidSynth (for the fluidsynth engine): brew install fluid-synth / apt install fluidsynth, plus a SoundFont (.sf2/.sf3). Point audiate at one with soundfont=... or $AUDIATE_SOUNDFONT; good free options include MuseScore_General.sf3 (MIT) and FluidR3_GM (MIT).
• MuseScore (for the musescore engine): the desktop app's mscore binary. Set $AUDIATE_MUSESCORE if it isn't on your PATH.

Where it fits

audiate is the score → audio stage of a larger pipeline:

• search a score by metadata → (a search facade)
• audiate — render that score to "MIDI audio"
• arioso — turn that into AI-enhanced audio (arioso.enhance)

Chord-chart arrangement (chords → a backing band) is accompy's job; audiate focuses on rendering fully-specified symbolic scores.

License

MIT