THE VOICE ROOM
Your voice. New words.
Capture a little of your voice. Give it something new to say.
What would you like to say?
The first run downloads the voice model.
Ready for this device
Check memory and storage before loading the model.
Run a short benchmark with your selected voice for a device-specific estimate.
GPU support varies by model and device. A local CPU path is always available. Generation remains English-only.
Hear the next version of your voice.
Take history
0 TAKESCompare takes and restore their scripts. Each finished take is kept separately.
Your first completed take starts the collection.
The first load downloads about 159 MB of model and runtime files. A current desktop browser with free memory works best. Speed depends on your device.
Getting the best result
Start with a clean sample
Use your normal speaking voice, close to the microphone in a quiet room. Select one speaker with steady volume. A phone call, background music or strong echo makes it harder to match the voice.
Shape the script
Use punctuation and shorter sentences. You can select one passage and generate that part again, or compare complete takes. Voice similarity, pronunciation and pacing vary; listen before sharing. This version supports English generation.
Keep your work
Working scripts and profiles stay in this browser. Finished speech saves to the browser and, when connected, your Creative Studios folder in Google Drive. A saved project includes its selected reference audio; treat the project as a personal voice file. Browser data can be cleared by your device, so keep a downloaded or Drive copy.
A real voice model
Pocket TTS learns the sound of the selected reference and generates new speech locally. It does not use the browser’s predefined read-aloud voices. Generated speech should be identified as AI-generated when sharing could otherwise mislead listeners.
Model: Kyutai Pocket TTS, CC-BY-4.0 · ONNX conversion and browser inference: KevinAHM · Attribution & licenses