Meet Basic Pitch: Spotify’s Open Source Audio-to-MIDI Converter

Introduced Basic Pitch, an open-source audio-to-MIDI converter built using machine learning. The post details the technical challenges of music transcription (polyphony, pitch bend, instrument agnosticism) and presents the model's architecture, which uses a harmonic constant-Q transform, jointly models onsets/frames/multipitch, and employs a shallow network. It highlights the model's lightweight nature (<20MB memory, <17K parameters) and accuracy compared to other systems, with a focus on enabling creators.