How to make an AI voice-over
- Type or paste your script, up to 10,000 characters.
- Pick a voice and a speed, then press Generate voice. The first time takes longer while the voice model downloads.
- Listen to the result, then download it as MP3 for videos and social posts, or WAV for editing.
The voices
All 13 voices are from Kokoro, a small open-source text-to-speech model that sounds remarkably natural for its size. Heart and Bella are its best American English voices; Emma and George speak British English. The voices read English only, so other languages will come out with an English accent.
| Voice | Accent | Good for |
|---|---|---|
| Heart, Bella | US female | Voice-overs, explainers, ads |
| Nicole | US female, soft | Calm narration, meditation, ASMR-style clips |
| Sarah, Sky, Nova | US female | Variety for characters and dialogue |
| Michael, Fenrir, Puck | US male | Narration and characters |
| Emma, Isabella | UK female | British narration |
| George, Fable | UK male | Storytelling, documentaries |
Making better voice-overs
- Short sentences. The voice phrases them more naturally and the audio is easier to cut to your edit.
- Punctuation is timing. Commas add a short pause, full stops a longer one, and new paragraphs a clear break.
- Spell things out. Write “twenty twenty-six” rather than 2026 and “A I” if an abbreviation is read as a word.
- Adjust afterwards. Use change pitch or volume booster on the downloaded file, or change audio speed for fine timing.
For instant playback with the voices built into your browser, use text to speech.