Speech Generated on Your Device
Your text and audio are processed in your browser and are not uploaded for synthesis. The model must be downloaded before first use.
Free · No account needed
Kokoro TTS — Free In-Browser Text-to-Speech
Kokoro TTS turns text into speech in 8 languages with 54 built-in voices, running entirely in your browser. No account, no upload, no per-character billing.
First use downloads the model. You can prepare your text while it loads.
0%English (US)75 / 2,000 characters
Your audio
—Write your text, choose a voice, and generate your first audio.
Generated on your device
Kokoro v1.0 82M
Why Kokoro TTS?
Text-to-speech in eight languages, generated on your device.
Free — No Signup Required
Kokoro TTS is free to use. Each input is limited to 2,000 characters; split longer scripts into sections.
WAV 24 kHz
The model runs locally after download. Pronunciation varies by language and voice; Japanese first use also downloads a pronunciation dictionary.
KOKORO VOICES
Find the voice for your words.
54 voices across 8 languages. Listen to native-language samples before you choose, from American English to Mandarin Chinese.
Explore all 54 voicesPerfect Kokoro Web Use Cases
From content creation to accessibility, Kokoro Web adapts to your voice needs
Content Creation
Narrate videos, podcasts, and multimedia projects with natural voices in English or any of the 8 supported languages.
Education & Learning
Convert study materials and e-learning content into audio so learners can listen along.
Accessibility
Transform written content into speech to improve access for visually impaired audiences.
Voice Applications
Power assistants and interactive apps with natural voice dialogue in English or another supported language.
KOKORO GUIDES
Kokoro TTS · Guides
Our articles focus on real tasks: choosing a browser, preparing scripts, exporting audio, checking quality, and recovering when a long render fails.
Explore Kokoro TTS GuidesGuides: setup, voice selection, and measured results
Hands-on write-ups from running this model in real browsers: which voices suit which language, why some pronunciations fail, and how the WASM and WebGPU paths compare in practice. Start with the browser setup guide, then the voice quality notes.
Every primary guide answers a concrete question, provides a repeatable process, and explains where the workflow can break.
About the SiteFrequently Asked Questions
Everything you need to know about free, private Kokoro TTS.
What is Kokoro TTS?
It is Kokoro v1.0, an 82M-parameter model built on the StyleTTS 2 architecture, running locally in your browser rather than on a server. On first use the browser downloads the model weights and attempts to cache them for reuse. Later sessions can reuse the cached weights, but still need to initialize the model; cleared or unavailable cache means another download. WebGPU is used where the browser supports it and WASM otherwise.
How does Kokoro TTS work in the browser?
The model runs on your device using WebGPU or WebAssembly, depending on browser support. The model is downloaded on first use and may need to be downloaded again if the browser cache is cleared.
Is Kokoro TTS really free? What’s the catch?
Yes, it is free to use without an account. Each input is limited to 2,000 characters.
Do I need to create an account?
No. No signup, login, or email is required. Open the page and start typing. That is it.
What voices does Kokoro TTS support?
Choose from 54 voices: English, Mandarin Chinese, Japanese, Spanish, French, Hindi, Italian, and Brazilian Portuguese. Russian and Korean interface translations do not add speech voices.
How does Kokoro TTS compare to cloud TTS services?
Cloud TTS services such as ElevenLabs, Google Cloud, and Azure render speech on their own servers. That means your text leaves your device, and usage beyond the free allowance requires a paid plan or usage-based billing, typically measured in characters or credits. Kokoro Web runs the 82M model inside your browser instead: nothing is uploaded for synthesis, there is no account, and there is no per-character billing.
The trade-offs are concrete. The model has to download on first use (WebGPU where available, WASM otherwise). Very long texts need to be split into segments. And there is no voice cloning or custom voice training. You choose from the 54 voices built into Kokoro v1.0.