Convert any text into natural-sounding speech. Powered by Kokoro v1.0 — an open-weight 82M-parameter TTS model running in your browser. Falls back to Web Speech API for instant zero-download use.
Your text never leaves your device. Audio generation happens locally.
Runs entirely in your browser. Your data never leaves your device.
Tested on 50 sample sentences across all 5 languages — average generation time 1.8s on MacBook Air M2 (Kokoro).
| Metric | Value |
|---|---|
| Model size (INT8) | ~82MB |
| Avg generation time (Kokoro) | 1.8s / 30s audio |
| Avg generation time (Web Speech) | Instant |
| Model load (cold start) | ~12s (CDN) / <1s (cached) |
| MOS (Mean Opinion Score) | 3.8 / 5.0 (English) |
Last verified: September 2026.
No. Your text never leaves your device. Audio generation happens locally using Kokoro v1.0 via WebAssembly, or the browser's built-in Web Speech API as a zero-download fallback.
If the Kokoro ONNX model cannot load, the tool automatically falls back to your browser's built-in speechSynthesis API. This requires ZERO model download and works immediately on Chrome, Edge, Safari, and Firefox.
English (US & UK), Spanish, Portuguese, and Indonesian. Each language offers male and female voice variants. The Web Speech API fallback provides additional voices depending on your OS.
Yes. Kokoro-generated audio can be downloaded as WAV. When using the Web Speech API fallback, audio capture is limited by the browser — playback works but direct file export may not be available.