New AI Clones Any Voice 10x Faster — No Script Needed

A team of researchers has unveiled Tacit-TTS, a voice-cloning system that can copy someone's voice from a short recording and then say anything you type — in that voice. The big deal is speed. Older systems build speech one sound at a time, like dictating a letter word by word. Tacit-TTS fills in the whole sentence at once, the way you'd complete a crossword puzzle by guessing every square simultaneously. The result: speech that arrives more than 10 times faster for clips longer than five seconds.

The second trick is that Tacit-TTS doesn't need a written transcript of the sample voice. Until now, most cloning tools required you to tell them exactly what the reference recording said, which meant running speech-to-text software first — and that step breaks down with accents, whispered speech, or unusual sounds. Because Tacit-TTS skips it, the system can clone voices from eight other languages, from a baby's babbling, even from nonsense gibberish. That's a real step toward cloning a voice from almost any clip you have lying around.

What's the catch? This is a research paper, not a product you can download today. It was built by compressing an earlier system, IndexTTS2, meaning it inherited that system's building blocks and its blind spots. Clone quality still depends on how clean and long your sample is. And the same convenience that helps creators also helps scammers: tools that copy any voice from a few seconds of audio, without knowing what the original said, make convincing voice fraud cheaper and easier.

Still, the direction is clear. Voice cloning is getting faster, lighter, and less fussy about the input it needs. Expect dubbed videos in your own voice, audiobooks narrated by you, and voice assistants that sound like a specific person. It also means you should treat an unexpected phone call from a loved one asking for money with more suspicion than you did last year.