- Shipped
- September 28, 2026 at 10:26 PM UTC
- Author
- Kamo
- Commit
- 84838bb
Kamo Speech (Speaches) answers 500 'No clip timestamps found' for a chunk of 30 s or more in which its VAD finds no speech (faster-whisper's no-VAD path refuses audio that long), and the whole recording failed. Live (SP14 Task 31, Step 9): a meeting's AI thought for 80 s in silence, the 30.016 s chunk cut through it failed twice, and the meeting kept no recording transcript. A chunk that one silence covers whole (ffmpeg silencedetect, the same log the cuts come from) is no longer sent at all and leaves no text, cue or segment; and a text chunk is at most 29.5 s, so a chunk of noise without speech takes Speaches' short-audio path instead of its error.