basic-pitch

Why we included this project

Musicians, producers, and audio developers who want a recording turned into editable notation or a playable MIDI file will get a working transcription tool with minimal setup. It runs a lightweight neural network from Spotify's Audio Intelligence Lab that converts polyphonic audio into MIDI with pitch bends, and it does not care which instrument you record, so a guitar solo or a vocal take comes out as notes. The package installs quickly with pip and includes pre-converted CoreML, TensorFlow Lite, and ONNX runtimes, so the default setup stays light on any platform. A command-line tool processes single files or batches, while the predict() interface lets you load the model once and run it over many tracks, which helps when transcription is part of a bigger pipeline. One honest limit: it handles one instrument at a time rather than separating a full mixed ensemble.