What AI model powers the text-to-speech?

This tool uses Kokoro-82M v1.0, an open-source neural TTS model by onnx-community built on the Supertonic architecture. It runs entirely in your browser via @huggingface/transformers (transformers.js) using WebAssembly and ONNX Runtime. Your text is never sent to any server — all processing happens on your own device.

Why is the first generation slow?

The Kokoro-82M model (~88 MB in q8 quantized form) downloads from Hugging Face on your first use. After that initial download it is cached permanently in your browser, so all subsequent generations — even after closing and reopening the page — start instantly without any download.

What voices and languages are supported?

The tool currently supports 28 English voices: 11 American English female voices (Heart, Alloy, Aoede, Bella, Jessica, Kore, Nicole, Nova, River, Sarah, Sky), 8 American English male voices (Adam, Echo, Eric, Fenrir, Liam, Michael, Onyx, Puck), 4 British English female voices (Alice, Emma, Isabella, Lily), and 4 British English male voices (Daniel, Fable, George, Lewis). All voices support the full speed range of 0.5× to 2.0×.

Free AI Text to Speech Online — Natural Voice

Convert any text to natural-sounding speech using Kokoro-82M AI, running entirely in your browser. 28 voices across American and British English accents, adjustable speed. Free, no signup, no upload.

How it works

1Type or paste up to 500 characters of text into the input box. The tool works best with natural prose, sentences, and short paragraphs.
2Choose a voice from 28 options — American and British English in male and female voices. Adjust the speech speed from 0.5× (slow) to 2.0× (fast). On first use, the Kokoro-82M model (~88 MB) downloads and caches permanently in your browser.
3Click "Generate Speech" — the Kokoro AI runs locally on your device and produces natural-sounding audio. Listen with the inline player, then download as a WAV file with one click.