New "Chunk Transcription" Mode

SaschaFuksa commented

2026-04-20 18:05:10 +00:00

Owner

Add a boolean flag in settings.yaml to enable Chunk Transcription Mode.
Default: False
Add overlap_time to settings, also activ in this mode.
Default: 5 seconds
If this Mode is true, chunks will get transcripted directly after they are finished. Flow:

First chunk is finished: Use chunk time like in config
Chunk will be stored to folder (current behaviour)
Chunk will now directly converted to *.wav
New *.wav will directly transcripted via background task
Transcripted text will be stored as a part_nnn.txt, example first text: part_000.txt, second as part_001.txt ...
6***n. Now next (second) chunk will get overlap! Add last 5 seconds of prev chunk. This is needed to preverese context for transcription.
After recording is stop, last chunk will be generated, send to folder an will be transkripted
After all transcriptions are finished, merge all *.txt to transcript.txt
In this mode, there is no need to create an full_audio.wav
This Feature shall be in an own PR! Don't batch this feature with other issues because it's a bigger change.

Add a boolean flag in settings.yaml to enable Chunk Transcription Mode. Default: False Add overlap_time to settings, also activ in this mode. Default: 5 seconds If this Mode is true, chunks will get transcripted directly after they are finished. Flow: 1. First chunk is finished: Use chunk time like in config 2. Chunk will be stored to folder (current behaviour) 3. Chunk will now directly converted to *.wav 4. New *.wav will directly transcripted via background task 5. Transcripted text will be stored as a part_nnn.txt, example first text: part_000.txt, second as part_001.txt ... 6***n. Now next (second) chunk will get overlap! Add last 5 seconds of prev chunk. This is needed to preverese context for transcription. 7. After recording is stop, last chunk will be generated, send to folder an will be transkripted 8. After all transcriptions are finished, merge all *.txt to transcript.txt 9. In this mode, there is no need to create an full_audio.wav This Feature shall be in an own PR! Don't batch this feature with other issues because it's a bigger change.