Both run 100% in your browser (no server for speech or matching) and use hold-to-talk. Test each with real speech, then tell me which feels right.
① Vosk — offline, streaming, grammar-locked to catalog ② Whisper — offline (WebGPU), best accuracy, Tamil| Vosk | Whisper | |
|---|---|---|
| Live text while holding | ✅ yes | ❌ after release |
| Speed | instant | ~1s (GPU) |
| Tamil words | ❌ EN only | ✅ yes |
| Accuracy on your catalog | ✅ locked to it | very good |
| First-load download | 36 MB | ~80 MB |
Use Chrome/Edge on desktop. First load downloads the model (then cached). Allow the microphone.