- Shipped
- November 7, 2026 at 12:04 AM UTC
- Author
- Kamo
- Commit
- 5693129
Owner 2026-11-06: "any time you have separate audio channels, make sure the transcripts leverage that so it can additionally say who is speaking." Every recorded channel was one mono MixMonitor mix of both sides, so no recording held one party alone (measured: every track of a multi-track call transcribes to the same words). [kamo-record] now also writes MixMonitor's r() and t() files, {stem}.rx.wav (what the channel said: that leg's own party) and {stem}.tx.wav (what it heard), beside the unchanged {stem}.wav. The grouper keeps them in their call and leg, and the manifest tags them "direction":"rx"|"tx". ConversionService (3ff1cc4) transcribes each side on its own, names it (our member by extension, else the other party), merges the turns in time order, and keeps the files as extra tracks; it never mixes them into the master.