Transcribe therapy session recordings to formatted text. Converts audio to clean, speaker-labeled transcripts (Me/Therapist format) with grammar correction and English translation. Use when processing therapy recordings, session audio, or any two-person conversation recording.
Part of the mindwork suite. Converts therapy session recordings into clean, formatted transcripts.
OPENAI_API_KEY environment variable setmindwork-transcribe Docker image built (see Setup)Build the transcription Docker image from the plugin's transcribe directory:
# Find the mindwork plugin location and build the image
docker build -t mindwork-transcribe ~/src/mindwork/transcribe
Or if installed as a plugin, find the plugin path first:
# The transcribe tool is in the 'transcribe/' directory of this plugin
docker build -t mindwork-transcribe /path/to/mindwork/transcribe
Transcribe, format as conversation, and translate to English:
docker run --rm \
-e OPENAI_API_KEY \
-v $(pwd):/data \
mindwork-transcribe /data/session.m4a --format-conversation --output /data/transcript.txt
Just transcribe without formatting or translation:
docker run --rm \
-e OPENAI_API_KEY \
-v $(pwd):/data \
mindwork-transcribe /data/session.m4a --output /data/transcript.txt
For automatic speaker detection (alternative to --format-conversation):
docker run --rm \
-e OPENAI_API_KEY \
-v $(pwd):/data \
mindwork-transcribe /data/session.m4a --diarize --output /data/transcript.txt
Split a large file into chunks for later processing:
docker run --rm \
-v $(pwd):/data \
mindwork-transcribe /data/session.m4a --no-transcribe --keep-chunks
Resume from previously created chunks:
docker run --rm \
-e OPENAI_API_KEY \
-v $(pwd):/data \
mindwork-transcribe /data/chunks/ --format-conversation --output /data/transcript.txt
| Option | Description |
|---|---|
--output FILE | Save transcript to file (default: stdout) |
--format-conversation | Format as Me/Therapist dialogue + translate to English |
--diarize | Auto-detect speakers (uses gpt-4o-transcribe-diarize) |
--no-transcribe | Only chunk, skip transcription |
--keep-chunks | Preserve chunk files after processing |
--model MODEL | whisper-1 (default, fast) or gpt-4o-transcribe (better accuracy) |
mp3, mp4, m4a, wav, webm, ogg, flac
If a mindwork.yaml config file exists, use it to determine output paths: