In the ListenBrainz project we've created a number of tools and techniques that allowed us to create the MBID Mapping sub-project. This project takes an artist name and a recording (track) name and attempts to find the best matching recording in MusicBrainz.
This library will contain code to facilitate matching non-MusicBrainz metadata to MusicBrainz (tag a music collection) and from MusicBrainz to non-MusicBrainz metadata (content resolution).
One component in this toolkit is the metadata cleaner, which removes (often) useless gunk from a metadata recording string:
"Tabula Rasa (feat. Lorraine Weiss)" -> "Tabula Rasa"
"Don't Give up - 2001 remaster" -> "Don't Give up"
This library also supports artist name cleaning:
"Amy Winhouse, Lorraine Weiss" -> "Amy Winehouse"
When a piece of metadata fails to be identified the metadata should be run through the metadata cleaner in an effort to remove this useless gunk that might be preventing a match. If the cleaner sucessfully removes gunk, the cleaned metadata should be looked up again, hoping that a match will be found this time.
To use this class, instantiate the MetadataCleaner class and call the clean_recording or clean_artist methods:
from lb_matching_tools.cleaner import MetadataCleaner
mc = MetadataCleaner()
mc.clean_recording("Tabula Rasa (feat. Lorraine Weiss)")
mc.clean_artist("Amy Winhouse, Lorraine Weiss")
To install this python module, simply do:
pip install .
Create virtual environment
python -m venv .venv
Install dependencies
pip install .
To run the internal unit tests, run:
pip install .
./test.py