Sonic Annotator: A Beginner’s Guide to Audio Feature ExtractionAudio feature extraction turns raw sound into measurable information — pitch, loudness, timbre, rhythm — that programs can analyze. Sonic Annotator is a lightweight, command-line tool designed to run audio analysis plugins (Vamp plugins) over audio files and produce time-aligned feature data. This guide explains what Sonic Annotator does, how it works, why it’s useful, and how to get started extracting features from audio.
What is Sonic Annotator?
Sonic Annotator is an open-source command-line application that applies Vamp plugins to audio files and exports the analysis as structured annotation files (CSV, JSON, RDF, etc.). It doesn’t perform the analysis itself; instead, it serves as a host and automation layer for Vamp plugins, which implement specific feature-extraction algorithms (e.g., spectral centroid, beat tracking, chroma).
Key facts:
- Sonic Annotator hosts Vamp plugins to extract audio features.
- It outputs annotations in formats such as CSV, JSON, and RDF.
Why use Sonic Annotator?
Sonic Annotator is useful when you need to process large audio collections reproducibly and efficiently. It is especially suited for researchers and developers working on music information retrieval (MIR), audio content analysis, dataset creation, or machine learning pipelines.
Benefits:
- Batch processing of many files from the command line or scripts.
- Supports a wide range of Vamp plugins for different tasks.
- Outputs machine-friendly formats that integrate with analysis pipelines.
- Lightweight and scriptable — no GUI required.
Core concepts
- Vamp plugins: analysis modules implementing algorithms (installed separately). Each plugin exposes “features” that Sonic Annotator can request.
- Feature output: can be time-stamped events, continuous timeseries, or global descriptors (file-level values).
- Labeling/annotation formats: Sonic Annotator supports several exporters (e.g., CSV, JSON, RDF), enabling downstream use in Python, R, or other tools.
- Batch processing: process directories or entire datasets with a single command.
Installing Sonic Annotator
Sonic Annotator runs on Linux, macOS, and Windows (via binaries or package managers). Installation steps differ by platform; these high-level commands cover the common cases:
-
On Debian/Ubuntu:
sudo apt-get update sudo apt-get install sonic-annotator -
On macOS (Homebrew):
brew install sonic-annotator -
On Windows: Download a prebuilt binary from the official project releases or use MSYS/WSL to run the Linux version.
You also need Vamp plugins. Popular choices include the vamp-plugins package and the QM Vamp plugins (e.g., for vamp-plugins + qm-vamp-plugins).
Install common Vamp plugins on Debian/Ubuntu:
sudo apt-get install vamp-plugin-sdk vamp-plugin-host vamp-plugins Or on macOS:
brew install vamp-plugins Finding and listing plugins
To see which Vamp plugins Sonic Annotator can use, run:
sonic-annotator -l This prints available plugins and their feature outputs. Each plugin is identified by an ID (e.g., vamp-example:example) and lists features and parameters.
Basic usage examples
- Run a single plugin and output CSV:
sonic-annotator -d vamp:qm-vamp-plugins:qm-barbeattracker:beats -w csv -r input.wav
- -d specifies a plugin descriptor (plugin ID and feature).
- -w chooses an output writer (csv).
- -r processes a single file or directory (recursive with -r).
-
Extract multiple features at once:
sonic-annotator -d vamp:qm-vamp-plugins:qm-vamp-plugins:tempo -d vamp:mtg-melodia:melodia:melody -w csv input_folder/(Use
-donce per plugin/feature; check plugin IDs withsonic-annotator -l.) -
Output JSON:
sonic-annotator -d vamp:qm-vamp-plugins:qm-barbeattracker:beats -w jann input.wavCommon writers include:
- csv — simple tabular output.
- jann — JSON annotation format.
- label — simple label files.
- rdfs — RDF-based output for semantic applications.
Understanding plugin parameters
Many Vamp plugins expose parameters (e.g., window size, hop size, threshold). When you list plugins with sonic-annotator -l, parameters are shown. To specify parameter values, include them after the plugin descriptor like so:
sonic-annotator -d vamp:plugin:id:feature:parameter=value -w csv input.wav Example adjusting minimum beat confidence:
sonic-annotator -d vamp:qm-vamp-plugins:qm-barbeattracker:beats:threshold=0.5 -w csv input.wav (Note: exact syntax and supported parameters vary by plugin.)
Common feature types and example plugins
- Beat and tempo: qm-vamp-plugins (barbeat/tempo detectors)
- Melody and pitch: mtg-melodia (melodic line extraction), qm-vamp-plugins pitch detectors
- Chroma (harmonic content): vamp-plugins chroma extractors
- Spectral features: spectral centroid, flux, rolloff from various plugin suites
- Onset detection: onset detectors in many plugin sets
Example: extract melody and beats for dataset labeling:
sonic-annotator -d vamp:mtg-melodia:melodia:melody -d vamp:qm-vamp-plugins:qm-barbeattracker:beats -w csv my_song.wav Working with outputs in Python
CSV and JSON outputs are easy to load. Example using Python (pandas) for CSV:
import pandas as pd df = pd.read_csv('my_song.csv') # Inspect timestamps and values print(df.head()) For jann (JSON) use Python’s json module to parse timelines and extract timeseries or events.
Tips for robust processing
- Test plugins on a few sample files before batch-running large datasets.
- Choose appropriate sample rates and window/hop sizes for your analysis goals.
- Redirect outputs to per-file directories to avoid overwriting when processing many files.
- Be mindful of plugin dependencies and install the matching Vamp plugin packages for your platform.
Troubleshooting
- “Plugin not found”: ensure the Vamp plugin is installed and listed by
sonic-annotator -l. - Wrong feature names: check the exact feature IDs in the plugin listing.
- Output empty files: try different plugin parameters or verify input audio format and duration.
- Permission or codec issues: convert audio to WAV or a supported PCM format using ffmpeg:
ffmpeg -i input.mp3 -ar 44100 -ac 1 output.wav
Example workflow: creating a labeled dataset
- Collect audio files into a folder.
- Decide which features you need (tempo, beats, melody, chroma).
- Use sonic-annotator with the chosen plugins, outputting CSV or JSON into a results folder.
- Post-process the annotations into your dataset format (e.g., align frames, aggregate features per track).
- Train or evaluate models using the extracted features.
Command example processing a dataset recursively and saving CSVs per file:
sonic-annotator -t -d vamp:qm-vamp-plugins:qm-barbeattracker:beats -w csv -r /path/to/audio -o /path/to/output (Use -t for per-file output naming; check version flags for exact options.)
Alternatives and complementary tools
- Essentia — a C++/Python library offering many built-in features and a Python API for richer programmatic control.
- Librosa — Python library for audio and music analysis; excellent for prototyping and feature extraction in Python.
- Vamp plugins + Sonic Visualiser — Sonic Visualiser provides a GUI for inspecting plugin outputs; Sonic Annotator is the headless batch counterpart.
Comparison (short):
| Tool | Best for |
|---|---|
| Sonic Annotator | Batch processing with Vamp plugins |
| Essentia | Rich feature set + Python/C++ integration |
| Librosa | Prototyping and research in Python |
Further reading and resources
- Sonic Annotator documentation and plugin lists (consult the project’s docs on your platform).
- Vamp plugin repositories for specific feature implementations.
- Tutorials on building MIR datasets and using extracted features for machine learning.
Sonic Annotator is a practical, scriptable bridge between Vamp feature implementations and real-world data workflows. For anyone preparing audio datasets, running large-scale analyses, or integrating established audio feature extractors into automated pipelines, Sonic Annotator is an efficient, reproducible choice.
Leave a Reply