libsonare

Turn audio into data and back, from Python. Analyze songs (BPM, key, chords, loudness), master and mix to broadcast loudness, and render MIDI through built-in instruments — a fast C++ core with NumPy as its only dependency.

Mastering ships 76 named DSP processors implemented against published references (ITU-R BS.1770-4 true-peak limiting, Linkwitz-Riley crossovers, Vicanek matched-Z biquads, ADAA-antialiased saturation); analysis defaults match librosa where the two overlap (validated against generated librosa reference values in CI). Apache-2.0, no model weights.

📖 Full API reference, guides, and CLI docs: libsonare.libraz.net

Installation

pip install libsonare

Supported platforms: Linux (x86_64, aarch64), macOS (Apple Silicon).

Quick Start

Audio is the recommended entry point: it decodes files and caches samples. The top-level libsonare.detect_* / libsonare.analyze functions are thin wrappers for one-shot calls on a numpy array.

import libsonare

audio = libsonare.Audio.from_file("song.mp3")  # or "song.wav"
result = audio.analyze()  # BPM + key + time signature + beats
print(f"BPM: {result.bpm:.1f}  Key: {result.key.root.name} {result.key.mode.name}")

# Master toward a target loudness with a named preset
mastered = libsonare.master_audio(
    audio.data, sample_rate=audio.sample_rate, preset_name="streaming",
)
print(mastered.output_lufs, mastered.applied_gain_db)

Analyze a numpy array directly (mono float32; downmix stereo first):

import numpy as np

samples = np.asarray(my_mono_float32_signal, dtype=np.float32)
bpm = libsonare.detect_bpm(samples, sample_rate=22050)
key = libsonare.detect_key(samples, sample_rate=22050)  # Key(root, mode, confidence)

Render a MIDI arrangement through a built-in instrument with the headless Project (a context manager):

with libsonare.Project() as project:
    project.set_sample_rate(48000)
    _, clip_id = project.add_midi_clip(0.0, 4.0)
    project.set_midi_events(clip_id, [
        libsonare.Project.midi_note_on(0.0, 0, 0, 60, 100),  # ppq, group, channel, note, velocity
        libsonare.Project.midi_note_off(2.0, 0, 0, 60),
    ])
    audio = project.bounce_with_synth_instrument("saw-lead", num_channels=2)

Capabilities

Every area below has runnable examples and the full API in the documentation. The functional and Audio-method forms return identical results; Audio caches decoded samples and is preferred when doing more than one computation on the same signal.

• Analysis — BPM, key (+ candidates), chords, downbeats, sections, melody, tuning; pitch (YIN / pYIN), timbre, and the full spectral feature set (STFT, mel, MFCC, chroma, CQT/VQT, spectral contrast); metering (metering_*, waveform_peaks). → Python API
• Mastering — 76 named DSP processors, the configurable mastering_chain, 25 named presets via master_audio, dynamics / repair specialist functions, and reference-matching. → Mastering processors
• Mixing — offline mix_stereo and the block-based Mixer with scene presets. → Mixing
• Editing DSP — time-stretch, pitch-shift, HPSS (+ residual), phase vocoder, normalize, trim, remix. → Editing DSP
• Room acoustics — blind RT60 / EDT, impulse-response clarity metrics, estimate_room, synthesize_rir, room_morph. → Room acoustics
• Realtime & streaming — RealtimeEngine (transport / MIDI / render / capture), StreamAnalyzer, StreamingMasteringChain, RealtimeVoiceChanger. → Realtime & streaming
• Instruments & synthesis — built-in oscillator synth, patch-driven NativeSynth (15 synthesis engines, incl. physically-modeled piano / strings / winds — being tuned over time), and a GS-compatible SoundFont (SF2) player. → Python API
• Headless DAW — Project arrangement model: audio / MIDI tracks and clips, undo/redo, SMF / MIDI 2.0 Clip File I/O, deterministic JSON, offline bounce. → Python API
• Conversions — Hz / mel / MIDI / note, frames / time, resample.

Native return-code failures raise libsonare.SonareError (a RuntimeError subclass carrying a numeric .code); input-validation failures (empty / NaN / Inf buffers, bad shapes) raise ValueError.

CLI

The sonare command exposes the analysis, mastering, mixing, effects, and project surfaces. A few representative commands:

sonare analyze song.mp3                              # BPM + key summary
sonare bpm song.mp3 --json                           # {"bpm": 161.0}
sonare master song.wav -o mastered.wav --preset pop  # preset mastering
sonare voice-change vocal.wav -o out.wav --preset bright-idol
sonare project bounce --in project.json -o out.wav --synth saw-lead

Run sonare --help (or sonare <command> --help), or see the CLI reference for the full command list.

Supported audio formats

The PyPI wheels are pinned to SONARE_WITH_FFMPEG=OFF so the distributed wheel never silently links against the build host's FFmpeg. To enable FFmpeg-backed decoding, build from source with SONARE_FFMPEG=1 (see the installation guide); this links against the system FFmpeg shared libraries (LGPL by default), so install them first (brew install ffmpeg, or apt install libavformat-dev libavcodec-dev libavutil-dev libswresample-dev).

Input format expectations

Stereo input passed as samples is not downmixed automatically — downmix yourself (e.g. samples.mean(axis=1, dtype=np.float32)). File loaders downmix to mono internally.

librosa-compatible defaults

Documentation

Full API reference and guides live at libsonare.libraz.net (getting started · Python API · CLI).

Also available

npm install @libraz/libsonare  # JavaScript / TypeScript (WASM)

License

Apache-2.0