Transcription
Cadeo transcribes your recordings into searchable, timestamped text. You can choose between local (on-device) and cloud transcription.
Local Transcription
Runs entirely on your device — no data leaves your machine.
Web (Whisper.js)
Uses OpenAI's Whisper model via your browser. Choose a model based on your needs:
| Model | Size | Speed | Quality |
|---|---|---|---|
| Tiny | ~40 MB | Fastest | Good |
| Base | ~75 MB | Fast | Better |
| Small | ~150 MB | Moderate | Great |
| Medium | ~500 MB | Slower | Best |
The first use downloads the model in the background.
Desktop
On macOS 26 or later, Cadeo uses Apple's on-device speech recognition. It handles long meetings and silences cleanly, runs in seconds and needs no model download from Cadeo. macOS downloads its speech model once, on first use. Speakers are still identified by Cadeo.
On earlier versions of macOS, or if Apple's recognition isn't available for your language, Cadeo uses its own Whisper models (sherpa-onnx) with built-in speaker identification:
| Model | Size |
|---|---|
| Tiny | ~117 MB |
| Base | ~200 MB |
| Small | ~374 MB |
Download models in Settings > Recording > Transcription. On macOS 26 or later you don't need to download one.
Cloud Transcription (Web Only)
Uses AssemblyAI for fast, high-accuracy transcription with automatic speaker identification.
- Faster processing than local
- Higher accuracy especially for accents and technical terms
- Speaker identification included by default
- Requires internet connection; recording data is sent to AssemblyAI servers
Choosing Your Transcription Method
Go to Settings > Recording and select:
- Local — for privacy and offline use
- Cloud — for speed and accuracy
Transcription Progress
During transcription, a progress indicator shows the percentage complete. Once finished, the full transcript appears with:
- Timestamps for each segment
- Speaker labels (e.g., "Speaker 1", "Speaker 2")
- Searchable text — find specific topics or quotes
Re-Transcribing
Click Generate Transcript on any recording to re-transcribe it, for example after switching to a different model or method.