Updated · Practical guide
Kokoro TTS in Your Browser: Setup, Voice Selection, and Measured Results
Kokoro Web uses the Kokoro-82M model to turn text into a WAV file in your browser. Follow the export workflow, then use the same English script to compare model setup and generation across browser runtimes.
What Kokoro Web currently does
The site loads the Kokoro-82M v1.0 model into a browser worker. When WebGPU is available, the app selects its WebGPU runtime; otherwise it uses a WebAssembly path. The model files must be fetched before the first generation and may be cached in browser storage for later sessions. Cache behavior and startup time vary by browser, connection, and available storage.
Once the model is ready, the entered text is passed to the local worker for synthesis. This means the text does not need to be uploaded to a speech-generation server. It does not mean the entire website is network-free: the page, model files, and separate site services can still make requests. The current tool accepts up to 2,000 characters and offers 54 voices across English, Mandarin Chinese, Japanese, Spanish, French, Hindi, Italian, and Brazilian Portuguese. Choose a voice matching the text language and export generated speech as WAV. Japanese first use also downloads a pronunciation dictionary.
Generate and check a WAV file
- Open the Kokoro TTS tool. The editor and voice settings are available before the model loads.
- Paste up to 2,000 characters of English text. Use normal spelling and punctuation. The current text box does not interpret SSML tags or bracketed acting notes as controls.
- Choose a voice and a speed between 0.5× and 2.0×, then select Generate. On first use the model loads and the requested audio generates automatically. The status shows whether the session selected WebGPU or WASM. Listen for names, dates, abbreviations, and sentence breaks.
- Download the WAV and make exact pauses, loudness adjustments, or edits in an audio editor when the project needs them.
For a long script, split at natural topic or sentence boundaries so each part stays within the input limit. There is no measured evidence here for a universal “best” segment duration, loudness target, or voice-to-content formula, so choose those settings by listening to the intended material.
WebGPU and WASM with the same script
On November 9, 2026, Kokoro Web ran its local static preview on one Mac Studio with an Apple M1 Max, 32 GB of RAM, and macOS 14.8.7. Each run used the text below (421 characters), Heart (af_heart), English (US), and 1.0× speed. Chrome used its normal WebGPU path and, in a separate run, a local test build with WebGPU detection disabled. Firefox and Safari selected WASM automatically.
Before publishing a narration, read the text aloud once and listen for names, numbers, and sentence breaks. This sample uses ordinary English punctuation so the same input can be repeated on another device. The browser downloads the speech model when it is first needed, then runs this synthesis locally. Keep the original text, choose one voice and speed, and listen to the full WAV before editing the script for rhythm.
A local timer recorded the first Generate click, the transition from loading to generating, and the appearance of the completed player. Setup includes fetching or reading model assets and initializing the runtime. The generation phase can include loading the voice file. Each row is one first request in a fresh tool tab. Asset caches and network conditions were not reset or standardized between trials.
| Browser and runtime | Setup | Generating to player | Total | WAV length |
|---|---|---|---|---|
| Chrome 156.0.8078.81 · WebGPU FP32 | 35.60 | 11.16 | 46.77 | 27.78 |
| Chrome 156.0.8078.81 · WASM Q8 (forced) | 15.52 | 122.93 | 138.45 | 27.93 |
| Firefox 144.0 · WASM Q8 | 77.42 | 77.43 | 154.85 | 27.93 |
| Safari 17.6 · WASM Q8 | 48.57 | Still generating after five minutes of observation | ||
The WebGPU run on this Mac delivered the first file sooner despite spending more time on setup than the forced WASM run in Chrome. Model precision also changed, and some WASM trials overlapped, so the measurements cannot isolate the speed difference between the runtimes or rank browsers. In a separate October 29 run with the model already initialized, Chrome WebGPU generated a 27.78-second WAV in 6.33 seconds. The hardware model and exact browser build were not recorded for that earlier run.
Listen to two of the measured outputs
These are MP3 previews of the original 24 kHz mono WAVs. Both clips use the same text and voice. Listen for the pauses between sentences and check the ending before comparing how they would fit a video cut.
How to make a fair comparison
Text-to-speech tools differ in language, voice, speed controls, network path, and export format. A fair performance test holds the input, voice style, output settings, device, and network conditions steady; it reports both generation time and audio duration; and it repeats each run after clearly labeling cold and cached starts. A voice-quality comparison also needs a listening rubric and multiple listeners. The timing table records how long the local tool took to produce the files. To compare a different browser or service, repeat the script and settings and record your own start and stop points.
- Privacy: this app synthesizes text in the browser after the model loads. Check another provider’s own documentation for its upload and retention practices.
- Startup: Kokoro Web requires model files on first use; browser caching can reduce later loading, but is not guaranteed to persist.
- Language and controls: the current generator offers 54 voices across eight languages and plain-text controls. Check another tool’s supported languages and markup before choosing it.
- Voice quality: listen to the same representative script in the voices you would actually use. Parameter count alone does not establish which voice sounds better.
Limitations and related guidance
WebGPU availability depends on the browser, operating system, and graphics hardware. If the app selects WASM instead, measure a short sample on your own device before committing to a long script. Very long scripts may need several generations because of the 2,000-character input cap. The browser can also clear cached model files, so keep a network connection available when you need to reload them.
Use the Kokoro voice guide to review the available voices and the model author’s grading notes. For sound cleanup after export, see our audio quality guide. The privacy policy explains analytics, model-host, and advertising requests that are separate from speech synthesis.
Frequently asked questions
Does Kokoro Web work offline?
A first model download is required. Browser storage may retain the files, but it can be cleared and the site may still need to load other assets or services, so offline operation is not guaranteed.
Can I enter SSML, pauses, or emotion tags?
The tool accepts plain text. SSML and bracketed stage directions are not special commands. Use punctuation to guide phrasing and edit precise pauses after downloading.
Does this test compare WebGPU with WASM?
Yes. The same script was run with WebGPU and WASM on one Mac. The observed timings include different model precisions and some overlapping trials, so they do not isolate runtime speed.
Is the entered text sent to a speech server?
The current app runs synthesis in a browser worker on your device. The page and model still need to load, and the site may make separate analytics or advertising requests; see the privacy policy for those services.