Audio Transcription

Convert any audio element on the canvas into a transcript. The transcript appears as a note next to the audio.

When to use it

• YouTube audio — extracted with Right-click → Extract audio, transcribed for note-taking.
• Voice memos — dropped onto the canvas.
• Podcast clips — to pull quotes.
• Generated speech — to verify what was said.

How to run it

1. Select an audio element (MP3, M4A, WAV, OGG, FLAC) on the canvas.
2. Right-click → Transcribe.
3. The app sends the audio to the configured transcription provider.
4. The transcript drops onto the canvas as a Markdown note next to the source.

Provider

The transcription uses the first available:

• OpenAI Whisper (if an OpenAI key is set)
• Google Gemini (if a Google key is set)

The transcription endpoint accepts any language and auto-detects.

Output

The result is a single Markdown note. Long transcripts include speaker turns when the source has clear pauses and natural breaks.

Cost

Whisper and Gemini transcription cost a fraction of a cent per minute of audio.

Limits

• Maximum audio length depends on the provider — Whisper caps single requests at ~25 MB (about 25 minutes of compressed audio). Longer files are split automatically.
• Heavy background noise lowers accuracy. Pre-process the audio (denoise, normalise) for cleaner results.

Tip

For YouTube interviews, the fastest path is:

1. Paste the YouTube URL on the canvas.
2. Right-click the video → Extract audio.
3. Right-click the audio → Transcribe.

The transcript and the video stay together on the board, ready to be quoted into notes or summarised by the agent dock.

Where to go next

• Voice Input — dictate prompts.
• Notes and Markdown Blocks — work with transcript output.
• Agent Dock Overview — mention a transcript with @ to summarise.