Smart Transcription Splitting

[fredjan3]

Automatically break up long transcriptions based on detected silence and duration:

• Short pause (1s) → new line break within the same box.
• Long pause (2s) → new transcription box
• Recording exceeds 60s and a long or short pause is detected → new transcription box

[fredjan3]

@tobiaswu probably need your take on this. Any better idea of how to handle this?

[tobiaswu]

@tobiaswu probably need your take on this. Any better idea of how to handle this?

Technical feasible. We already use a Voice Activity Detection (VAD) model that we could leverage to detect silences better. Another idea is to run a small on-device NLP model that enables quick voice cues like "new line" (single \n), "paragraph" (double \n/new box) to achieve that formatting.