What AI can now do·Jan 5, 2023·Media

VALL-E clones a voice from 3 seconds of audio

Microsoft Research

3 secondsEnrollment audio needed for a voice clone

Microsoft's VALL-E synthesized personalized speech for an unseen speaker from a 3-second enrolled recording, outperforming prior zero-shot text-to-speech on naturalness and speaker similarity.

How far it has travelled

From a demonstration to real life, in five steps. Empty steps are the point: a demonstration is not a deployment.

Demonstratedshown once, by its developer or a researcher
yes · Jan 5, 2023 · arXiv paper (Jan 5, 2023): zero-shot TTS from a 3-second acoustic prompt; demo pages only, no model release. source
Reproducedshown again by someone else
yes · Jan 31, 2023 · ElevenLabs' January 2023 public beta cloned voices from 'a few short snippets'; celebrity-voice abuse via 4chan was reported by Jan 31, 2023. source
Accessibleavailable to the public or to customers
yes · Jun 30, 2023 · ElevenLabs offered a free tier at launch and cloning for paying users; more than 1 million registered users by June 2023; 28 languages by August 2023. source
Reliableworks most of the time on a defined task
yes · Feb 4, 2026 · Production voice cloning at scale for three years (company valued at $11B in Feb 2026); the company shipped an AI Speech Classifier in June 2023, but detection remains a separate, contested problem. source
Real consequenceshas changed something in the world, for good or ill
yes · Jan 22, 2024 · Family-emergency voice scams (DeStefano, 2023), Slovakia's election-eve audio (Sept 2023) and the Biden New Hampshire robocall, which was linked to ElevenLabs tools and drew a $6M FCC fine. source

What this could change for you

Anyone's voice can be faked from a short clip, so a familiar voice on the phone is no longer proof of who is calling.

Incidents it made possible

Sources

primary source, second source.

A milestone is a dated, sourced result that moved what AI can do; a stage is upgraded only when a source shows it (a reproduction, a public release, production evidence, a documented consequence). “Unknown” is the honest default.