transcribing-meeting-recordingslisted

# Transcribing Meeting Recordings (with Speaker Identification) ## Overview End-to-end pipeline: video/audio file → cleaned SRT transcript with timestamps and **real participant names**. Uses **whisperx** (faster-whisper + alignment + pyannote diarization) for ASR + speaker clustering, plus a Microsoft Teams / Zoom **video-frame trick** to map anonymized `SPEAKER_XX` clusters onto real names. ## When to Use - User provides a meeting recording (`.mp4` / `.mov` / `.mkv` / `.wav` / `.m4a` / `.mp3`) and wants a transcript with timestamps. - User asks for SRT/VTT subtitles, captions, 字幕, or "transcript with speaker labels". - Multilingual content (especially Chinese with English technical terms mixed in — whisper large-v3 handles this well with the right prompt). - User wants to know **who said what** in a meeting, not just the words. **Do NOT use for:** - Live / real-time / streaming transcription — whisperx is batch-only. - Summarization without producing a transcript artifact. - Audio-only files where speaker identification doesn't matter — then skip diarization and the Teams-frame step. ## Quick Reference | Stage | Command | | ------------------------------ | ---------------------------------------------------------------------------------------- | | 1. Extract WAV | `ffmpeg -i in.mp4 -vn -ac 1 -ar 16000 -c:a pcm_s16le audio.wav`