Split out from the second half of #60, which fixed only the reported bug (MP3 output leaving the WAVs behind).
@hugochisholm also raised:
I wonder whether it might be worth having a setting to explicitly delete OR save the system and mic audio as separate files following a recording. I think in general the final combined version is "good enough," but there may be value for some users who may wish to retain those separate component files for future processing.
Reasonable. Every recording currently keeps three tracks — mic_audio, system_audio, combined_audio — and for most users only the combined one is ever opened. Dropping the components after the combined track is assembled would cut disk use per recording by roughly two thirds.
What has to be checked before building this, because the component tracks aren't purely archival:
SimpleDiarizer reads both mic_audio and system_audio to label You vs Remote — see .claude/rules/transcription-pipeline.md. Deleting them breaks simple diarization on any re-run, which is the no-setup default path for users without a HuggingFace token.- Re-running diarization on an existing recording (full or simple) would be permanently unavailable for pruned recordings.
So this can't be a plain "delete after stop" toggle. Options, roughly in order of how much they cost:
- Setting, default off, and disable/warn when simple diarization is the active mode.
- Delete components only after transcription and diarization have both completed for that recording, so the one consumer that needs them has already run.
- Keep components until the user explicitly prunes, via a per-recording or bulk "free up space" action in the recordings list.
(2) is the most honest fit for the pipeline but adds cross-component lifecycle coupling that doesn't exist today. (3) is the least invasive and keeps the destructive step user-initiated.
No strong preference yet — wanted the constraint written down before anyone starts.
Split out from the second half of #60, which fixed only the reported bug (MP3 output leaving the WAVs behind).
@hugochisholm also raised:
Reasonable. Every recording currently keeps three tracks —
mic_audio,system_audio,combined_audio— and for most users only the combined one is ever opened. Dropping the components after the combined track is assembled would cut disk use per recording by roughly two thirds.What has to be checked before building this, because the component tracks aren't purely archival:
SimpleDiarizerreads bothmic_audioandsystem_audioto label You vs Remote — see.claude/rules/transcription-pipeline.md. Deleting them breaks simple diarization on any re-run, which is the no-setup default path for users without a HuggingFace token.So this can't be a plain "delete after stop" toggle. Options, roughly in order of how much they cost:
(2) is the most honest fit for the pipeline but adds cross-component lifecycle coupling that doesn't exist today. (3) is the least invasive and keeps the destructive step user-initiated.
No strong preference yet — wanted the constraint written down before anyone starts.