Whistle Transcription Online
On-device ASR · 16.9MB · Privacy-first · vs Whisper
Speech-to-text with Cactus Whistle (16.9MB). Runs on-device in the browser. Not Whisper. Upload a file or record from the mic. Get a transcript and word timestamps on this page.
Transcribe on-device
Runs Cactus Whistle in this tab. Audio stays on your device. First load downloads ~18 MB (cached after).
Loading engine…
Whistle vs Whisper
| Aspect | Whistle | Whisper |
|---|---|---|
| Model size | ~16.9MB (Cactus Whistle) | Hundreds of MB to multi-GB |
| Typical deploy | On-device / browser WASM | Cloud or heavy local GPU |
| Privacy | Audio stays on device | Often API / large local runtime |
| Languages | 7 (en de fr es it nl pl) | 99+ |
| Best for | Lightweight demos, mobile, offline | Highest accuracy, rich languages |
FAQ
- What is Whistle Transcription?
- Whistle is a tiny on-device speech-to-text model from Cactus Compute (~16.9 MB). This page runs it in your browser via the needle WASM engine. Audio stays on your device.
- Is this Whisper?
- No. Whistle is a separate, much smaller model for on-device inference. It is not OpenAI Whisper.
- Does my audio leave the device?
- No. The model and engine download once from Hugging Face (~18 MB). Transcription runs in this tab. Your audio is not uploaded to our servers.
- Which languages work?
- English, German, French, Spanish, Italian, Dutch, and Polish. Other languages return empty or weak transcripts — use a broader model for those.
- Where is the model?
- See the Hugging Face model card: https://huggingface.co/Cactus-Compute/whistle