Chevron7Locked/kima-hub v1.12.1

Kima v1.12.1

on GitHub

A hotfix for 1.12.0. Nothing to do when upgrading: enrichment and your library carry on as they are.

The iPhone app is with Apple for testing. If you have an Apple developer account, you can build it and put it on your phone now: kima-ios.

Fixed

• Only the newest release's "what's new" window. A new account, or someone who skipped versions, was shown every release's window one after another (1.10, then 1.11, then 1.12). Now the window shows only the newest release. What an older release still asks of you (1.11's "wipe and re-run" for libraries enriched before it, and its saved mood mix question) is counted on the version number as open and can be done from the changelog page, which now also has the mood mix choice.
• A song whose file is gone is removed instead of holding enrichment up. When audio analysis, sound vectors or fingerprinting find no file for a song, it is set aside at once and a library scan confirms it and removes it, with its album and artist when they are left empty (logged first, as scans do). It used to be retried for over an hour, holding the moods and similarity back. Nothing is removed when the whole music folder is missing (a drive that is not mounted), and a file that is there after all goes back into enrichment.
• A restart no longer stalls audio analysis. Songs the audio analyzer was working on when it restarted (a container restart, a crash, or turning the GPU switch on or off) went back to the server only after 15 minutes, and since sound vectors wait for every song's analysis, a restart near the end of a run held everything up for that long. They now go back in the queue within about a minute and are analysed once.
• Enrichment runs in order on a fresh install. Enrichment used to start on the songs a library scan had added so far, so on a new server audio analysis and sound vectors took turns batch by batch while the scan was still running, and the moods and similarity were rebuilt from part of the library every few seconds. Now enrichment waits until the scan has finished. Identification, artist metadata and audio analysis run side by side, then the sound vectors for every song, then the moods, then similarity. Audio analysis and sound vectors each finish every song they owe before the other starts, unless "Run audio analysis and sound vectors at the same time" is on; a song added in the meantime waits its turn. Mood tags, mood mixes and stations, and similarity now appear once the sound vectors are done for the whole library, not song by song.

Quick start

docker run -d \
  --name kima \
  -p 3030:3030 \
  -v /path/to/your/music:/music \
  -v kima_data:/data \
  chevron7locked/kima:v1.12.1

Then open http://localhost:3030 and create your account. See the README for full documentation.

Full Changelog: v1.12.0...v1.12.1