Basic Pitch
Lightweight instrument-agnostic audio-to-MIDI converter with pitch bend detection, from Spotify.
About
Spotify's Audio Intelligence Lab designed Basic Pitch to make automatic music transcription trivially cheap: the instrument-agnostic model converts audio to MIDI with pitch bend detection using a neural network small and fast enough to run on CPU, and the research behind it was published at ICASSP 2022. It handles polyphonic and multipitch material across instruments including voice and guitar, and writes standard MIDI files ready for any DAW. The Python package installs with pip install basic-pitch on macOS, Windows, and Ubuntu, needs no GPU, and ships the model in TensorFlow, CoreML, TensorFlow Lite, and ONNX formats, defaulting to the best option per platform; a sibling TypeScript port runs the same model in the browser, powering the free hosted demo at basicpitch.spotify.com. Both a command line tool and a Python API are included. Apache-2.0 licensing covers the code and model, so commercial products can embed it. With about 5,400 GitHub stars, it has become the default open tool musicians and music information retrieval researchers reach for when recorded audio needs to become editable notes.
Should you use Basic Pitch?
Pick it when
Pick Basic Pitch when you need recorded audio turned into editable MIDI with pitch bends on ordinary hardware: it runs on CPU, handles polyphonic voice or guitar, embeds commercially under Apache-2.0, and runs in the browser.
Look elsewhere when
Skip it as a one-step answer for full mixes: the authors say it works best on one instrument at a time, so split stems with Demucs first. For beats, tempo, and spectral features, librosa is the better tool.
Alternatives to Basic Pitch
- librosa
ISC-licensed analysis library for beat tracking, tempo, and spectral features, but it gives you signals to work with rather than ready MIDI notes.
- Demucs
Not a transcriber, but splitting vocals, bass, and drums first lets Basic Pitch transcribe one part at a time from a busy mix.
- Spleeter
Faster stem splitting with 2, 4, or 5 stem models to run before transcription, though TensorFlow on Apple Silicon can be awkward.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Category
- Music & Audio Generation
- Price
- Free
- Platform
- Hybrid
- Difficulty
- Easy (2/5)
- License
- Apache-2.0
- Added
- Jul 29, 2026