long-audio-transcription

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

1. Probe source audio with ffprobe.
2. Convert to 16 kHz mono PCM WAV with ffmpeg.
3. Apply optional speech enhancement and loudness normalization.
4. Split long recordings into 5-10 minute chunks with 3-5 second overlap.
5. Transcribe chunks using faster-whisper or openai-whisper.
6. Restore original global timestamps.
7. Remove repeated text in overlap regions.
8. Apply only explicit terminology corrections.
9. 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.