bindings for Chromaprint acoustic fingerprinting and the Acoustid API
Chromaprint and its associated Acoustid Web service make up a high-quality, open-source acoustic fingerprinting system. This package provides Python bindings for both the fingerprinting algorithm library, which is written in C but portable, and the Web service, which provides fingerprint lookups.
This library works with Python 2 (2.7+, possibly also 2.6) and Python 3 (3.3+).
First, install the Chromaprint fingerprinting library by Lukas Lalinsky. (The library itself depends on an FFT library, but it’s smart enough to use an algorithm from software you probably already have installed; see the Chromaprint page for details.) This module can use either the Chromaprint dynamic library or the fpcalc command-line tool, which itself depends on libavcodec. If you use fpcalc, either ensure that it is on your $PATH or set the FPCALC environment variable to its location.
$ pip install pyacoustid
You can run the included demonstration script, aidmatch.py, to test your installation:
$ python aidmatch.py mysterious_music.mp3
This will show the top metadata match from Acoustid’s database. The script uses audioread to decode music, so it should transparently use a media library available on your system (GStreamer, FFmpeg, MAD, or Core Audio).
Using in Your Code
The simplest way to use pyacoustid to identify audio files is to call the match function:
>>> import acoustid >>> for score, recording_id, title, artist in acoustid.match(apikey, path): >>> ...
This convenience function uses audioread to decode audio and parses the response for you, pulling out the most important track metadata. It returns in iterable over tuples of relevant information. Everything happens in one fell swoop. There are also a number of “smaller” functions you can use to perform parts of the process:
- fingerprint(samplerate, channels, pcmiter): Generate a fingerprint for raw audio data. Specify the audio parameters and give an iterable containing blocks of PCM data.
- fingerprint_file(path): Using either the Chromaprint dynamic library or the fpcalc command-line tool, fingerprint an audio file. Returns a pair consisting of the file’s duration and its fingerprint.
- lookup(apikey, fingerprint, duration): Make a request to the Acoustid API to look up the fingerprint returned by the previous function. An API key is required, as is the length, in seconds, of the source audio. Returns a parsed JSON response.
- parse_lookup_result(data): Given a parsed JSON response, return an iterator over tuples containing the match score (a float between 0 and 1), the MusicBrainz recording ID, title, and artist name for each match.
The module internally performs thread-safe API rate limiting to 3 queries per second whenever the Web API is called, in accordance with the Web service documentation.
If you’re running your own Acoustid database server, you can set the base URL for all API calls with the set_base_url function.
Calls to the library can raise AcoustidError exceptions of two subtypes: FingerprintGenerationError and WebServiceError. Catch these exceptions if you want to proceed when audio can’t be decoded or no match is found on the server. NoBackendError, a subclass of FingerprintGenerationError, is used when the Chromaprint library or fpcalc command-line tool cannot be found.
- Include fpcalc.py script in source distributions. Add Python 3 support (thanks to Igor Tsarev).
- Include fpcalc.py, a script mimicking the fpcalc program from the Chromaprint package. Handle a UnicodeDecodeError raised when using the fpcalc backend on Windows with Unicode filenames. Standard error output from fpcalc is suppressed.
- Properly encode Unicode parameters (resolves a UnicodeEncodeError in fingerprint submission). Parse all recordings for each Acoustid lookup result.
- Add a new function, fingerprint_file, that automatically selects a backend for fingerprinting a single file.
- Fix response parsing when recording has no artists or title. Fix compatibility with Python < 2.7. Add specific NoBackendError exception.
- Fingerprinting can now fall back to using the fpcalc command-line tool instead of the Chromaprint dynamic library so the library can be used with the binary distributions (thanks to Lukas Lalinsky). Fingerprint submission (thanks to Alastair Porter). Data chunks can now be buffers as well as bytestrings (fixes compatibility with pymad).
- Configurable API base URL. Result parser now generates all results instead of returning just one. Find the chromaprint library on Cygwin. New module names: chromaprint and acoustid (no package).
- Compress HTTP requests and responses. Limit audio decoding to 120 seconds. Return score from convenience function.
- Initial release.
This library is by Adrian Sampson. Chromaprint and Acoustid are by Lukas Lalinsky. This package includes the original ctypes-based bindings written by Lukas. The entire library is made available under the MIT license. pyacoustid was written to be used with beets, which you should probably check out.