Tavlo AI Lab / speech & audio ai
Speech & Audio AI
Transcribe your voice with OpenAI Whisper and hear text spoken aloud — all on-device.
on-device · audio never leaves your device
Tip: click Record, speak a sentence, then click again to stop. The first run downloads the Whisper model once.
The first Speak downloads the ~92 MB Kokoro voice once, then it's cached. “Browser built-in” speaks instantly with no download.
Try clapping, whistling, snapping, or playing music. The first run downloads the audio model once, then it classifies on your device every ~1.5 s.
Private by design. Transcription uses OpenAI Whisper running locally via Transformers.js — your microphone is accessed only after you press Record, and your audio is transcribed on your device and never uploaded, recorded to disk, or stored. Text-to-speech uses the open-source Kokoro-82M neural voice model, also running locally (~92 MB one-time download; a browser built-in voice is available as an instant fallback). See our privacy policy for details.