Whisper in-browser speed test
Everything runs locally in this browser tab. Models download once from Hugging Face and are cached; no audio leaves this machine.
1. Environment
Detecting…
2. Audio
No audio loaded yet.
3. Runs
| Model | Backend | Clip | Load s | Transcribe s | Audio s | Speed | 1 min audio ≈ | WER | Gate | Transcript |
|---|
Gates: base·GPU must do 1 min of audio in < 20 s · tiny·CPU in < 60 s. WER = word error rate vs the JFK reference (or vs the best GPU run for other clips). Lower is better.