텍스트 음성 변환
Neural AI voice (Kokoro) and browser voices · all running locally in your browser.
About This Tool
Neural Voice (Kokoro AI) uses a state-of-the-art 82-million-parameter neural text-to-speech model. It runs 100% in your browser using WebAssembly and ONNX Runtime · no text is sent to any server. The first time you use it, the model (~100 MB) downloads and is cached by your browser. After that, it loads instantly. You get 13 voices across American and British accents, male and female, each with natural intonation and prosody. Generated audio can be played back and downloaded as a WAV file.
Browser Voices use your system's built-in Web Speech API. They're instant with zero download, but voice quality and availability depend on your OS and browser. Chrome typically offers the most voices. Great for quick previews and accessibility testing.
Frequently Asked Questions
Does this upload my text to a server?
No. Both modes run entirely in your browser. The neural model is downloaded once and runs locally via WebAssembly. No text or audio data leaves your device.
How large is the neural model download?
Approximately 100 MB on first use. Your browser caches it automatically, so subsequent visits load the model instantly with no re-download.
Which browsers support the neural voice?
Any modern browser with WebAssembly support · Chrome, Firefox, Edge, and Safari. Chrome and Edge tend to perform best. A device with at least 4 GB RAM is recommended.
What is Kokoro?
Kokoro is an open-source neural TTS model with 82 million parameters, producing natural-sounding speech. We use the quantized ONNX version for efficient in-browser inference.