worldveil/dejavu

Dejavu is a Python library that can identify a song by listening to a short clip of it, similar in concept to the way music recognition apps work on a phone. It does this by analyzing audio to create a compact "fingerprint" and storing those fingerprints in a database. When you play an audio clip or hold a microphone up to a speaker, Dejavu compares what it hears against the stored fingerprints and tells you which song matches. The README notes it works well for recognizing exact recordings with background noise, but is not built for voice or speech recognition. The recognition process has two stages. First, you run a fingerprinting step, which processes a folder of audio files and saves their fingerprints to a database (MySQL or PostgreSQL are both supported). This can take time for large collections, but the process is resumable: if you stop and restart it, Dejavu picks up where it left off. Second, you run a recognition step against either a saved audio file or live microphone input. Both a Python API and a command-line interface are available for both stages. Setup is available through Docker for the quickest start, which brings up both the Python environment and a PostgreSQL database together. There is also a manual installation path described in a separate file for people who prefer to configure their own database. The configuration is passed as a plain Python dictionary, which includes the database connection details and a few optional parameters like how many seconds of each song to fingerprint. Adjustable settings in the config file let you tune the tradeoff between recognition speed and accuracy, with documentation on what each parameter does. The README includes test result plots showing how confidence scores and recognition speed change as the fingerprint database grows. The project appears to be in maintenance mode based on the README's age, but the Docker setup and examples remain functional. The README does not specify a license.