transcribe
Transcribe audio files from meetings into text documents using Whisper. Use when the user types /transcribe, has a new audio recording, or when RA detects new audio files in meetings/audio/. Supports speaker diarization with pyannote.
When & Why to Use This Skill
This Claude skill automates the conversion of audio recordings into structured text documents using the state-of-the-art Whisper model. It features advanced speaker diarization via pyannote, allowing users to transform meetings, seminars, and interviews into searchable, speaker-labeled markdown files. By integrating seamlessly with downstream summarization tools, it streamlines the transition from raw audio to actionable insights.
Use Cases
- Meeting Documentation: Automatically transcribe business meetings to create a permanent record and facilitate the extraction of action items.
- Research & Interviews: Convert academic seminars or qualitative research interviews into text for easier analysis and citation.
- Automated File Processing: Monitor specific directories to automatically detect and transcribe new audio uploads without manual intervention.
- Multi-Speaker Analysis: Utilize speaker diarization to distinguish between different participants in complex discussions, ensuring clarity in the final transcript.
| name | transcribe |
|---|---|
| description | Transcribe audio files from meetings into text documents using Whisper. Use when the user types /transcribe, has a new audio recording, or when RA detects new audio files in meetings/audio/. Supports speaker diarization with pyannote. |
Audio Transcription
Transcribe audio files from meetings into text documents.
Usage
/transcribe [filename]
/transcribe .research/meetings/audio/2024-12-02-lab-meeting.m4a
/transcribe .research/meetings/audio/ # Transcribe all untranscribed audio in directory
When to Use
- After recording a meeting, seminar, or discussion
- When RA detects new audio files in meetings/audio/ folder
- Before running /summarize_meeting
Supported Formats
- .m4a, .mp3, .wav, .webm, .mp4 (audio track)
- .ogg, .flac
Execution
The command runs:
conda run -n research-assistant python .ra/skills/transcribe/scripts/transcribe.py [filename or .research/meetings/audio/]
Behavior:
- If
[filename]provided: transcribe that specific audio file - If no filename (or
.research/meetings/audio/specified): automatically detect all audio files without transcripts and process them - If transcript already exists for a file: skip it
- Output saves to
.research/meetings/transcripts/[same-name].md
Post-Transcription Options
Transcription complete!
A) Run /summarize_meeting to extract action items and create tasks
B) Open transcript to review manually first
C) Continue with other work
What would you like to do?
Quality Notes
Improving Transcription Quality
- Use good microphone/recording quality
- Minimize background noise
- Speak clearly and at moderate pace
- Identify speakers at start if possible
Limitations
- Speaker diarization may be imperfect
- Technical terms may need manual correction
- Timestamps are approximate
Configuration
Environment variables (optional):
WHISPER_MODEL: Model size (default: "small", options: tiny, base, small, medium, large-v3)WHISPER_LANGUAGE: Force language (default: auto-detect)HF_TOKEN: HuggingFace token for speaker diarization
Related Skills
summarize-meeting- Extract action items from transcriptnext- Get next suggestion
Notes
- Raw transcripts may contain errors - review before citing
- Keep original audio files as source of truth
- Transcripts are for internal use, not publication