💛 Donate to help keep AudioMuse-AI growing. Your support covers AI subscription for development (~€220/month), homelab electricity for testing (~€20/month), and makes the project sustainable.
Release Date: September 4th, 2026
AudioMuse AI v3.6.0 introduces a new Neural Fingerprint model that powers Search by Recording: hold your phone up, record about 20 seconds of whatever is playing, and AudioMuse AI picks out the exact song in your library and highlights it for you, then lines up a list of similar tracks to keep the vibe going. Not in your library? You still get the closest similar matches, so the search never comes up empty. No tags, no metadata, just the audio itself. Prefer a file? Upload a clip instead and get the same answer.
The Neural Fingerprint model ships disabled by default. To turn it on, open the Setup Wizard and meet the new Machine Learning Models tab: every model in one place, with its analysis status, an on/off switch, and a plain-language explanation of what it does and which feature it powers. Enable it, run the analysis, and the Search by Recording page comes alive.
This release also brings a serious rework of the queue behind the batch tasks: instant cancellation, self-cleaning task rows, and acknowledged service restarts, so long analyses stop the moment you ask and a restart no longer leaves stale state behind. The legacy AI database user has been removed as well, which means one fewer role to manage and one fewer default credential to worry about.
Important
Browsers only hand the microphone to pages served over HTTPS or from localhost. If you already run AudioMuse-AI behind a proxy with a valid certificate, you are set. If not, the app answers HTTPS on the same port it already binds, using a self-signed certificate it creates on first start, so there is no extra port to publish and no config change: accept the browser warning once and Record works. Uploading a clip also works over plain HTTP, if you would rather skip the warning entirely.
Note
Turning on the Neural Fingerprint model adds the new Search by Recording page to the menu. Your library is not fingerprinted yet: each song picks up its fingerprint during the next analysis run, so run one after enabling it.
What's Changed
- Queue Centralization by @NeptuneHub in #892
- Queue centralization improvement by @NeptuneHub in #895
- Search track with Neural Fingerprint by @NeptuneHub in #897
- AI DB USER REMOVED - #898 by @NeptuneHub in #903
Full Changelog: v3.5.2...v3.6.0