[App] Subgen - AI subtitle generation for your *arr-stack

Share your Projects!
1

SubGen runs Whisper locally to automatically generate subtitles for your media library, designed to work alongside Bazarr as a fallback when no external subtitles can be found. This uses a LOT of RAM for storing the movie and CPU for the AI model, so it will probably not work well on a RPi.
The full movie will be loaded up in RAM, and therefor a 2 hour movie will easily take up 8+ Gb of RAM.

Features:

  • Local Whisper-based transcription, running on your HA host
  • Bazarr integration: timeouts and score thresholds are configurable so SubGen only kicks in when Bazarr’s own providers come up empty
  • Works alongside an existing Jellyfin/Sonarr/Radarr/Bazarr/Prowlarr stack

Installation:

  1. Add https://github.com/Xornop/ha-apps as a repository under Settings → Apps → Install app → ⋮ → Repositories
    or just use this Open your Home Assistant instance and show the add app repository dialog with a specific repository URL pre-filled.
  2. Install “Subgen”
  3. Configure the Bazarr connection in the app options: provider settings, score threshold, and timeout
  4. Restart Bazarr so it picks up the new provider
2

cool. ive not looked at the repo, but in relation to the comment about RAM use when looking at movies…is it possible to initially extract the audio only with ffmpeg?

ffmpeg -i input.mp4 -vn -acodec copy output.aac

maybe that helps.

Cool stuff

3

Thanks for replying. Your idea seems great, however, Bazarr already extracts the audio before sending it to Subgen/WhisperAI. The RAM issue is with both the AI model and loading the entire file, and mainly cause by Subgen. Since this is just an App wrapper, I probably can’t improve the RAM use much, but you’re welcome to make a PR on the git