| name | long-audio-transcription |
| description | Process long audio recordings with normalization, overlapping chunking, Whisper transcription, global timestamp restoration, duplicate removal, and domain terminology prompts. |
Long Audio Transcription
Workflow
- Probe source audio with ffprobe.
- Convert to 16 kHz mono PCM WAV with ffmpeg.
- Apply optional speech enhancement and loudness normalization.
- Split long recordings into 5-10 minute chunks with 3-5 second overlap.
- Transcribe chunks using faster-whisper or openai-whisper.
- Restore original global timestamps.
- Remove repeated text in overlap regions.
- Apply only explicit terminology corrections.
- Export TXT, SRT, JSON, and processing metadata.
Quality rules
- Keep raw transcription separate from corrected transcription.
- Never invent unclear speech.
- Mark uncertain sections for review.
- Preserve original timeline coverage.
Typical use
Suitable for meetings, defenses, interviews, lectures, and research discussions containing technical terminology.