Skip to content

AudioMuse AI v3.5.1: DCLAP Text Search Concept Weights

Latest

Choose a tag to compare

@NeptuneHub NeptuneHub released this 30 Aug 14:52
3775acc

💛 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: August 30, 2026

AudioMuse AI v3.5.1 improves DCLAP Text Search with a new Concept Weight functionality.

AudioMuse-AI aims to bring recent university research into your homelab. This release is based on the paper Steering Dense Music Retrieval with Open-Vocabulary Concept Discovery by Julien Guinot, Alain Riou, Elio Quinton, and György Fazekas.

Until now, you could search for something like:

- POP Viola with Female vocalist

The problem is that POP and Female vocalist can be very common concepts in a music collection, while Viola is much more specific. As a result, Viola can have less influence on the final results.

With the new Add Concept button, you can add a concept with a custom weight, for example:

  • Viola ×3

This gives Viola, an already present concept in your query, more importance in the search and helps steer the results towards songs where that concept is more relevant.

Note

Concept Weight works especially well when increasing the weight of a concept that is already present in the original query. However, it can also be used with concepts that are not in the query, or by reducing the importance of an existing concept. The final result depends on the query, the selected concepts, and your music collection.

This release also improves audio analysis for songs with corrupted segments. A song is now marked as not analyzable only when less than 50% of its audio segments are readable. This allows partially corrupted songs to be analyzed when enough valid audio is available.

Note

If you run the docker-compose standard example you can clean the not analyzable query list with the below query to give a second chance. Just remember to put your correct password and user if you changed.

docker exec -it audiomuse-postgres psql -U audiomuse -d audiomusedb \
  -c "DELETE FROM analysis_exclusions;"

What's Changed

Full Changelog: v3.5.0...v3.5.1