7 Practical Uses for Sonic Annotator in Music Analysis and Research

From WAV to Features: How to Run Sonic Annotator — Step-by-Step

Overview

Sonic Annotator is a command-line tool for batch extracting audio features from files using Vamp plugins. This guide shows a practical, minimal workflow: install, pick plugins, run extraction, and inspect results. Assumes Linux/macOS (Windows steps noted where different).

1) Install Sonic Annotator

• Linux (Debian/Ubuntu):
  • sudo apt update && sudo apt install sonic-annotator
• macOS (Homebrew):
  • brew install sonic-annotator
• Windows:
  • Use prebuilt binaries from the Sonic Annotator project or install via Cygwin/MSYS with package build; alternatively run in WSL.

2) Install Vamp plugins you need

• Common plugins:
  • QM-vamp-plugins (tonal, chroma)
  • vamp-libxtract
• Linux:
  • sudo apt install vamp-plugin-host qm-vamp-plugins vamp-plugin-sdk
• macOS (Homebrew):
  • brew install qm-vamp-plugins vamp-plugin-libxtract
• Confirm plugin availability:
  • sonic-annotator -l

3) Choose the plugin and output format

• Plugins have plugin:feature IDs shown by sonic-annotator -l.
• Output formats:
  • CSV (–csv) — good for spreadsheets, machine processing
  • RDF (–rdf) — semantic annotations
  • JSON (–json) — structured programmatic use
• Common choice for features: CSV or JSON.

4) Basic command examples

• Extract a single feature to CSV:
  • sonic-annotator -d vamp:qm-vamp-plugins:specparam:spec -w csv input.wav -o output.csv
• Extract multiple features:
  • sonic-annotator -d “vamp:qm-vamp-plugins:melbands:melbands,vamp:qm-vamp-plugins:chroma:chroma” -w csv input.wav -o features.csv
• Process a folder (recursive):
  • sonic-annotator -r -d vamp:qm-vamp-plugins:melbands:melbands -w csv /path/to/audio -o /path/to/output
• Use JSON:
  • sonic-annotator -d vamp:qm-vamp-plugins:tonal:tonal -w json track.wav -o tonal.json

5) Common useful flags

• -r : recurse directories
• -t : show plugin parameters and types
• –raw : include raw frames (if supported)
• –csv-stdout / –json-stdout : print to stdout for piping
• –tempo-method / plugin-specific params: append parameter settings after plugin ID (consult -t)

6) Inspecting and cleaning outputs

• CSV columns typically: file, start, duration, label, value
• Use Python (pandas) or R to load and aggregate per-frame features:
  • pandas.read_csv(‘features.csv’) then groupby file or time window
• Normalize or aggregate (mean, median, rms) over time windows to get fixed-size feature vectors.

7) Batch workflow example (practical)

1. Install sonic-annotator and qm-vamp-plugins.
2. List plugins: sonic-annotator -l > plugins.txt
3. Choose features (e.g., melbands, chroma).
4. Run recursively: sonic-annotator -r -d “vamp:qm-vamp-plugins:melbands:melbands,vamp:qm-vamp-plugins:chroma:chroma” -w csv /dataset/audio -o /dataset/features
5. Postprocess: use a script to compute per-file summaries and save as ARFF/CSV for ML.

8) Troubleshooting

• Plugin missing: check plugin ID with -l and ensure plugin package installed.
• Permission errors on Windows: run as admin or use WSL.
• Large outputs: use JSON streaming or split by file; adjust frame size via plugin parameters.

9) Tips

• Start with -l to explore available features.
• Run on a single file to tune plugin parameters before batch processing.
• Keep outputs organized mirroring input folder structure.
• Use –csv-stdout to pipe directly into processing scripts for speed.

If you want, I can:

• generate the exact sonic-annotator command for a specific feature set and OS,
• provide a short Python script to aggregate CSV outputs into per-file feature vectors.