Quality default
Try Kokoro-82M free — unlimited, in your browser
Kokoro is an 82-million-parameter StyleTTS2-derived model that consistently tops quality rankings among small open models. It is the best all-round choice: natural prosody, dozens of voices, and it still runs at real-time speed on CPUs without a GPU.
One-time download, stored in your browser and reused offline. Manage or remove it any time on the Models page.
Specifications
| Developer | hexgrad |
|---|---|
| Parameters | 82 million |
| Download size | 88 MB CPU / 310 MB WebGPU (q8 (WASM) / fp32 (WebGPU)) |
| Output | 24.0 kHz mono |
| Voices | 28 |
| Languages | English (US), English (UK) |
| Voice cloning | No |
| Streaming | Yes — audio starts before generation finishes |
| License | Apache-2.0 — Model weights are Apache-2.0. The phonemizer used by browser ports derives from espeak-ng (GPL-3.0). |
| Speed | ≈ real-time on a modern CPU (WASM); 2–6× faster than real-time with WebGPU. |
| Runtime | WebAssembly, WebGPU-accelerated when available |
Strengths
- +Highest English quality per megabyte of any browser model
- +48 distinct voices with blendable styles
- +Runs everywhere — WASM real-time, WebGPU fast
Limitations
- −English-focused (community Chinese variant exists)
- −No voice cloning
- −WebGPU uses the 310 MB fp32 file; CPU/WASM uses the 88 MB q8 file
Voices (28)
Compare
Kokoro-82M — questions
Is Kokoro-82M free to use?
Yes. The weights are Apache-2.0 and they run on your device. Model weights are Apache-2.0. The phonemizer used by browser ports derives from espeak-ng (GPL-3.0).
What hardware does Kokoro-82M need?
It runs on any modern CPU via WebAssembly at roughly real-time speed; with WebGPU it is several times faster. 4 GB RAM recommended.
Is my text private when using Kokoro-82M here?
Yes. The model executes inside your browser tab — the text you synthesize is never transmitted to a server.
Can I use Kokoro-82M output commercially?
Review the current model, voice, and source-dataset terms before commercial use. The catalog lists Apache-2.0 for this model and notes: Model weights are Apache-2.0. The phonemizer used by browser ports derives from espeak-ng (GPL-3.0).