Local speech generation, without an API key
The Kokoro 82M model performs speech synthesis in your browser with Transformers.js. The first use downloads the model weights and voice data; later sessions can reuse the browser cache. Generation runs on your device, so the script is not sent to Accelerated Logic or a text-to-speech API.
Make a video voiceover in a few steps
Open Decks Studio to create a local voiceover for a video timeline, paste your narration, and select a voice. Generate and preview the audio, then insert it as an editable voiceover clip at the current playhead. Export mixes active timeline voiceover, soundtrack, and sound-effect clips with the rendered video. If MP4 audio is unavailable in your browser, export falls back to WebM so the audio track is preserved.
Use the same voice engine from chat
In chat, choose Voiceover from the + menu or request a voiceover in your prompt. When the assistant finishes, Kokoro creates a local audio card on its reply with playback, voice selection, regeneration, and WAV download. Local Voice Studio remains available for pasted scripts.
Know what to expect
Kokoro currently offers English voice presets, with American and British styles. The model returns 24 kHz mono audio; the tool creates WAV files and limits each script to 12,000 characters. Model download and speech generation time depend on your connection, browser, and device.
Frequently asked questions
Does voice generation upload my script?
No. The browser downloads model files from Hugging Face, then synthesizes the speech on your device. The script and generated audio are not sent to an AI API.
How large is the first download?
About 100 MB of model files and voice data. Your browser caches the model weights for later sessions.
Can I add the audio to a video?
Yes. In Decks Studio, generate a local voiceover from the video timeline and insert the previewed audio as an editable timeline clip.
Which voices and formats are available?
The current model offers American and British English voices. Generated speech can be played and downloaded as a WAV file.