UitbreidenOS When to activate
- Creating transcripts from audio recordings
- Formatting transcripts with speaker labels and timestamps
- Generating accessible text versions for hearing-impaired listeners
- Creating blog post content from episode transcripts
- Building searchable text archives of episode content
When NOT to use
- For live captioning (real-time services needed)
- For translation of transcripts
- For audio editing (use audio-editor)
Instructions
- Transcribe audio. Use Whisper, Deepgram, or Rev for initial transcription. Target >95% accuracy.
- Add speaker labels. Format:
[HH:MM:SS] Speaker Name:for each speaker turn. - Insert timestamps. Every 5 minutes and at segment transitions for easy navigation.
- Clean up text. Remove filler words (um, uh, like), false starts, and repetitions. Keep natural tone.
- Add formatting. Bold key terms, italicize emphasis, add [laughs], [music], [sound effect] markers.
- Verify accuracy. Spot-check 10% of transcript against audio; correct proper nouns and technical terms.
- Export formats. Plain text (.txt), markdown (.md), SRT/VTT for captions, and HTML for web publishing.
Example
# Episode 42 Transcript: Scaling APIs to 10M Requests/Day
[00:00:00] **Host:** Welcome back to the show. Today I'm joined by Dr. Sarah Chen, VP Engineering at ScaleAI. Sarah, thanks for being here.
[00:00:12] **Dr. Chen:** Thanks for having me! Excited to talk about scaling.
[00:00:18] **Host:** So let's jump right in. You recently gave a talk at KubeCon about ML at scale. What was the core message?
[00:00:28] **Dr. Chen:** The core message was that *ML infrastructure is 80% plumbing*. [laughs] People think it's about models, but it's really about the data pipelines...
[00:05:00] --- Segment: Architecture Decisions ---