The model doesn't load or the download stops
- Check your connection. The first download is about 200 MB per language.
- Make sure your device has free storage space: browsers refuse to cache files when the disk is nearly full.
- Private or incognito windows may not keep the model between visits.
- Use a recent browser version. Very old browsers can't run the voice engine.
- Still stuck? Click Delete downloaded models in the Studio sidebar, reload and try again.
“Microphone access was blocked”
Your browser asked for permission and it was denied. Click the camera/microphone icon in the address bar, allow the microphone and try again. On macOS also check System Settings → Privacy & Security → Microphone. Alternatively, record with another app and upload the file.
“That clip is too short”
After silence is trimmed, at least a few seconds of speech must remain. Record 10–20 seconds of continuous speech.
The clone sounds noisy, echoey or muffled
The model reproduces your reference clip exactly, including its problems. Record again following the recording guide: a quiet, soft room and a steady microphone distance.
The clone doesn't sound like the speaker
- Use a longer clip (10–20 seconds) with only one speaker.
- Avoid clips with music, effects or phone-call quality.
- Make sure the language selected in the Studio matches the language you're generating.
Words are mispronounced
Spell out numbers, abbreviations and unusual names phonetically ("Nguyen" → "Win"). Add a comma before a difficult word to slow down the delivery.
Generation is slow
- Use the dedicated Studio page: it can use several processor cores and is usually 2–4× faster than the home page demo.
- Close heavy tabs and apps while generating.
- On phones, generate shorter texts.
My saved voices disappeared
Voices are stored in your browser on this device. Clearing site data, using a different browser or a private window means they aren't available. Keep your original recordings so you can re-create voices any time.