Spotify open-sources a small tool for audio-to-MIDI

The audio task

A musician who records an idea by singing, humming, or playing an instrument often wants that performance turned into editable notation or MIDI rather than a generated substitute, so it can be quantized, transposed, or rescored inside a DAW. Existing pitch-to-MIDI tools have often been limited to one instrument at a time, or to monophonic input such as a single melodic line, which fails on a strummed guitar chord or a layered performance.

What the documents show

Spotify's Basic Pitch site describes it as “a free audio-to-MIDI converter with pitch bend detection, built by Spotify,” that works on “any single instrument (piano, guitar, xylophone, you name it).” The company's engineering blog post, published 1 June 2022 by research scientist Rachel Bittner, states the model performs polyphonic transcription, detecting multiple simultaneous notes rather than only one at a time, and includes pitch-bend detection for techniques like vibrato and glissando. The post describes the model architecture as a shallow neural network with fewer than 17,000 parameters, small enough to run without specialized hardware, and states the tool is designed “for musicians, not just researchers,” as “a great starting point for transcriptions, instead of having to write all their musical ideas down manually from scratch.” Neither document publishes a benchmark comparing Basic Pitch's transcription accuracy against a human transcriber.

Rights status

The engineering blog states Basic Pitch “comes with a standard open source license and can be used for any purpose, including integrating it with other music production tools.” Spotify does not name a specific licence family on either page, but the stated scope, any purpose, including commercial integration, is broader than a restricted research licence. Because the tool transcribes a user's own performance into MIDI rather than generating new musical material, neither document raises a question about rights to the transcribed content itself; the output is a machine-readable version of audio the user already recorded.

What to check before you use it

This is an editorial checklist. Before building a production pipeline around Basic Pitch, confirm the exact licence text in the tool's own repository rather than relying on the summary language on the marketing site, since the blog does not name the licence; treat the polyphonic transcription as a first-pass sketch to correct by ear, since no accuracy benchmark is published; and check whether the version installed matches the one described in Spotify's 2022 announcement, since open-source tools can be forked or modified after release.

• What specific open-source licence governs the version of Basic Pitch actually being used?
• Has the generated MIDI transcription been checked by ear against the source recording before it is used as a session's backbone?
• Is the installed build the one Spotify documents, or a modified fork with different behavior?

Basic Pitch is a useful contrast case within this category: a free, openly licensed transcription tool that converts a musician's own performance into an editable form, rather than a commercial plugin generating new material. The distinction between transcribing and generating is worth keeping clear whenever a tool description uses the word AI.