Hum2Song (https://www.carlostoxtli.com/hum2song/, documentation: https://github.com/toxtli/hum2song) generates a musical accompaniment from a hummed or sung melody.

Works locally and does not transmit any data: processing and models run in the browser (Google Chrome recommended).
Templates and sounds are downloaded on the first load (100 MB).

In “Simple” mode, a 2-8 second recording triggers the build.
The “Advanced” mode shows the stages of generation: audio transcription into notes, music genre detection, instrumental adaptation, bass and drums generation, scale, chords and accompaniment. This mode accepts an imported audio file or a connected MIDI keyboard.

The result downloaded in MIDI format can be edited in a dedicated interface (https://www.carlostoxtli.com/daw/) with editing of tracks, synthesizers, BPM settings and WAV export.

Open source, free, accountless, and unlimited.

Tests from hummed voice in “Simple” mode (converted to mp3 for this article):

Tests from hummed voice in “Advanced” mode (converted to mp3 for this article) with initial recording: