sounddiff is a CLI tool for audio producers and developers to compare two audio files and see exactly what changed. It reports differences in loudness, spectral balance, timing, and potential issues like clipping and silence. Output comes as colored terminal text, structured JSON, or a self-contained HTML report.
New to audio? The sounddiff wiki has guides to LUFS, spectral analysis, clipping, and everything else sounddiff measures.
$ sounddiff mix-v3.wav mix-v4.wav
sounddiff: mix-v3.wav vs mix-v4.wav
Duration 3:42.108 → 3:42.108 (no change)
Sample Rate 48000 Hz → 48000 Hz (no change)
Channels stereo → stereo (no change)
Loudness (integrated)
LUFS -14.2 → -12.8 (+1.4 dB)
Peak dBTP -1.1 → -0.3 (+0.8 dB)
LRA 8.2 → 6.4 (-1.8 LU)
Spectral
Low (20-250 Hz) +0.8 dB avg
Mid (250-4k Hz) +0.3 dB avg
High (4k-20k Hz) +1.9 dB avg
Segments
0:00-1:12 similar (correlation: 0.97)
1:12-1:14 ADDED (new content, 2.1s)
1:14-3:42 similar (correlation: 0.98, shifted +2.1s)
Issues
⚠ Clipping detected in mix-v4.wav at 2:31.4 (3 samples)
pipx install sounddiffOr with pip: pip install sounddiff
Requires Python 3.10 or later. Supports WAV, FLAC, OGG, and AIFF natively. For MP3, AAC, WMA, and Opus, install ffmpeg. See docs/install.md for detailed setup.
Compare two files with colored terminal output:
sounddiff old-mix.wav new-mix.wavGet structured JSON for scripts and CI pipelines:
sounddiff old.wav new.wav --format jsonGenerate an HTML report:
sounddiff old.wav new.wav --format html -o report.htmlSee docs/usage.md for all options.
| Category | Measurements |
|---|---|
| Loudness | Integrated LUFS, true peak (dBTP), loudness range (LRA) per ITU-R BS.1770 |
| Spectral | Average energy per frequency band (low, mid, high) with configurable ranges |
| Temporal | Segment-level cross-correlation, added/removed/shifted section detection |
| Detection | Clipping events (timestamp, channel, sample count), silence regions |
| Metadata | Duration, sample rate, channels, bit depth, format |
Terminal is the default. Colored, grouped by category, designed to be read top to bottom. Uses rich for formatting.
JSON outputs the same data in a structured format. Pipe it to jq, parse it in Python, or use it in CI pipelines for automated regression testing.
HTML generates a self-contained report with inline styles. No external dependencies. Open it in any browser, share it with your team, or archive it alongside your session files.
sounddiff is written in Python with a modular architecture. Each analysis type (loudness, spectral, temporal, detection) lives in its own module with no cross-dependencies. The core orchestrator loads two audio files, runs all analyzers, and passes the results to a formatter.
| Dependency | Purpose |
|---|---|
| soundfile | Audio I/O via libsndfile |
| numpy | Array math, FFT, cross-correlation |
| scipy | Signal processing |
| pyloudnorm | ITU-R BS.1770 loudness measurement |
| click | CLI framework |
| rich | Terminal formatting |
| jinja2 | HTML report templates |
See docs/architecture.md for the full module breakdown and data flow.
- Installation - system dependencies, shell completions, ffmpeg setup
- Usage - CLI options and examples
- API Reference - using sounddiff as a Python library
- Architecture - module layout and design decisions
Contributions are welcome. See CONTRIBUTING.md for setup instructions and our development workflow.
The issue board has open work organized by milestone. Issues labeled good first issue are scoped for newcomers and have enough context to get started without deep DSP knowledge.
Report vulnerabilities to dev@systemblue.io. See SECURITY.md for our disclosure policy.
