Release Date: September 04, 2025
AudioMuse-AI v0.6.6-beta introduces the experimental Collection Sync feature. After agreeing to the Privacy Policy and signing in via OAuth (GitHub only), users can sync their analysis data with a centralized online database.
Opting in lets you leverage the community’s contributions, reducing the need to analyze every song locally, while supporting future feature development. If you prefer to keep your data private, simply don’t use Collection Sync—your data will remain local.
We’ve also set embeddings as the default for clustering. You can still override this via an environment variable or the front-end advanced view, as embeddings are the preferred approach moving forward.
Important: Angular distance is now standardized for both Song Path and Playlist From Similar Song in the default configuration. Please run an analysis (even one album is sufficient) to update the Voyager index before using these features.
Highlights
- Added Collection Sync to integrate with the PocketBase centralized database.
- New
/externalAPI path to fetch score and embedding data (#60). - Security improvement: front-end no longer passes API keys or passwords; these must be set via environment variables (#57).
- Initial support for proxy configuration (work in progress) (#59).
- Improved duplicate song removal using distance in addition to name filtering (#51).
- Default clustering now uses KMeans with embeddings and PCA max 199.
- Default distance metric for similar song and song path set to angular distance.
What's Changed
- Add podman deployment examples by @K3CAN in #45
- Add Nvidia GPU support for neural net inference by @ravenscroftj in #44
- fix: monkeypatch np.float_ to make clustering work again by @ravenscroftj in #47
- Pocketbase by @NeptuneHub in #58
- Devel -> Main by @NeptuneHub in #63
New Contributors
- @K3CAN made their first contribution in #45
- @ravenscroftj made their first contribution in #44
Full Changelog: v0.6.5-beta...v0.6.6-beta