The best free AI voices for video, and how to direct them

A few years ago, free text-to-speech sounded like a sat-nav. Today, open neural models produce narration that most viewers won't question — if you choose well and direct a little. Here's how, using the voices built into the AI voiceover video maker.

The two engines

Kokoro (English)

Kokoro-82M is a small but remarkably natural open model released under Apache-2.0. It handles intonation, emphasis and rhythm better than most models many times its size. It offers 20 American and 8 British voices. Start with Heart (warm female), Michael (calm male), Emma (British female) or Fable (British storyteller).

Piper (48 other languages)

Piper is a fast, lightweight neural voice system (MIT licensed) with over 150 voices across 48 languages. Quality varies by voice — those trained on more data, labelled High quality or Natural, sound best. Many languages have several voices to choose from.

Matching the voice to the job

• Explainers and tutorials: clear, mid-pitched, unhurried. Speed 0.95–1.0×.
• Ads and social: brighter, a little faster. Speed 1.05–1.15×.
• Stories and kids' content: warm and expressive. Speed 0.9×.
• Documentary: lower, measured. Speed 0.9–0.95×.

Always audition with your own script rather than a demo sentence — the Hear this voice button reads your first sentence for exactly this reason.

Directing an AI voice

You can't give an AI voice notes, but you can shape its performance through the text:

• Punctuation is pacing. A comma is a breath; a full stop is a pause; a question mark lifts the ending.
• Split long sentences. Long sentences flatten intonation. Two short sentences sound more alive than one long one.
• Spell it how it sounds. If a name or brand is mispronounced, spell it phonetically in the narration (the on-screen title can keep the real spelling).
• Write numbers as words when the reading is ambiguous: “twenty twenty-six”, not “2026”.
• Acronyms: write “A I” or “S-Q-L” to force letter-by-letter reading, or “sequel” if that's how your audience says it.

Multilingual videos

Because the visuals don't depend on language, one project can become many. Duplicate your project, paste the translated script, and the maker detects the new language and selects a matching voice. Titles and captions render in every script, including right-to-left Arabic and Urdu, Devanagari and Chinese.

For a language without an AI voice — or for the most authentic result in any language — record a native speaker scene by scene using My voice.

Why these voices are free

Cloud TTS services charge per character because they run on their GPUs. Kokoro and Piper are small enough to run in your browser, so the maker downloads the model once (20–92 MB) and generates speech on your device indefinitely. No credits, no character limits, and your script stays private.

When to use a human instead

AI voices are excellent for information. For emotional storytelling, comedy timing, or a brand that is built on a specific person, a human voice — yours or a professional's — still wins. The maker supports both, so you can prototype with AI and record the final version yourself.

Try every voice in the animated video maker, or browse them on the languages page.