Transcribe audio to subtitles when no source exists #207

Open
opened 2026-08-24 21:28:40 +01:00 by naps62-yolo · 0 comments
Owner

Deferred out of the Subtitles milestone deliberately; filed so the reasoning
survives.

When no provider has a subtitle in any language and the file carries no text
subtitle track, there is nothing to translate from and the file simply stays
unsubtitled in the missing-subtitles queue. Transcribing the audio would close
that last gap.

Cost of doing it: whisper means either a resident model or a subprocess loading
several hundred megabytes per run, against the constraint the whole project
exists for — one binary at 20-40 MB doing what five .NET services do at
600-900 MB. Reaching a remote whisper endpoint over the existing
remote-command or OpenAI-compatible backend shape would sidestep the memory
cost, and is the version worth considering first.

DESIGN.md §15 states this as a non-goal, so picking it up means amending §15
first.

Deferred out of the Subtitles milestone deliberately; filed so the reasoning survives. When no provider has a subtitle in any language and the file carries no text subtitle track, there is nothing to translate from and the file simply stays unsubtitled in the missing-subtitles queue. Transcribing the audio would close that last gap. Cost of doing it: whisper means either a resident model or a subprocess loading several hundred megabytes per run, against the constraint the whole project exists for — one binary at 20-40 MB doing what five .NET services do at 600-900 MB. Reaching a remote whisper endpoint over the existing remote-command or OpenAI-compatible backend shape would sidestep the memory cost, and is the version worth considering first. DESIGN.md §15 states this as a non-goal, so picking it up means amending §15 first.
naps62-yolo added the area/substype/featuredifficulty/hard labels 2026-08-24 21:28:40 +01:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: yolo/arr#207