Google's new voice model copies a voice from 30 seconds of audio.
Google released two Gemini 3.8 text-to-speech models on Wednesday in its developer API and AI Studio. They can build a voice from a written description, or copy a real one from a 30-second sample. Copying needs a recorded consent clip, and Google switched the feature off in Europe, India, Illinois and Texas.
Why it mattersA voice you can order from a prompt makes audiobooks and dubbing cheap, and impersonation cheaper. The places where Google turned copying off are the ones with strict voice and biometric laws.
Read in the app — free, in 9 languages
Related stories
An open model that never writes text took 19,000 GitHub stars.2026-09-23One hidden setting turned Meta's AI assistant into a Mac wiretap.
2026-09-23OpenAI answered Anthropic's price cut 90 minutes later.
2026-09-23Anthropic's cheaper new Claude took first on an independent index.
2026-09-23A model trained only on ancient Greek is restoring broken papyri.
2026-09-22