Gemini 3.8 TTS clones voices from 30 seconds

Google launched Gemini 3.8 Flash TTS and Flash-Lite TTS, with Flash TTS able to replicate a voice from a 30-second sample and a library of 2,000-plus production voices. Google’s launch announcement says replication requires a matching verbal-consent recording and generated audio carries SynthID watermarking and C2PA credentials; both models support more than 100 languages and dialects. Voice-agent and media teams can prototype custom voices without a fixed catalog, but should log rights evidence and verify consent and watermark behavior before putting cloned voices in customer-facing flows; Gemini Live’s context-cost analysis shows why audio capability still needs a full operating budget.