Gemini speech models add voice creation and script-level direction

Google’s September 23 Gemini 3.8 Flash TTS and Flash-Lite TTS announcement adds custom voice design, voice replication and delivery instructions for individual script lines. Flash emphasizes audio quality, while Flash-Lite prioritizes speed and cost. Replication requires a matching consent recording from the voice owner.

Brief updated 25 Sep 2026, 02:25 UTC · AI-written

Text-to-speech converts written scripts into generated audio. The new controls let developers specify characteristics such as accent, timbre and pacing, then direct delivery within a two-speaker scene using AI Studio’s audio workspace.

Google says generated audio includes SynthID watermarks. SiliconANGLE recounts the launch and the models’ different production priorities, but does not independently test consent checks or watermarking against misuse.

What remains unclear

The opened sources describe safeguards without independently establishing their robustness.