Google announced Gemini 3.8 Flash TTS and Flash-Lite TTS on September 23. Flash is aimed at designing distinctive voices and directing expressive dialogue; Flash-Lite is positioned for higher-volume dubbing and audio production. The announcement describes script-level control over pacing, emotion, pauses and conversational cues, plus single-script scenes with two voices.

Google Gemini audio keyword visual
Gemini audio visual from Google’s announcement.blog.google

The announcement card names both new speech models; Google positions them for voice design and directed dialogue.

The feature with the largest identity implications is voice replication. Google says a user can build a consistent voice profile from a 30-second sample, provided they have rights to use the voice. Google also says replication requires a matching verbal-consent recording and that generated audio carries SynthID watermarking and C2PA credentials. These describe Google’s stated safeguards; they do not independently establish how well consent checks prevent abuse or how reliably third parties can detect a watermark.

Google cites Hume AI and Voice Arena evaluations to claim leading results in voice design and human preference. Those are vendor-reported rankings. The separate RW-Voice-EQ research paper argues that text-to-speech quality has several distinct dimensions—including naturalness, expressiveness, identity stability and reliability—so a single leaderboard position cannot answer every production question. The paper supplies useful evaluation context, not an independent test of Gemini 3.8.

The release is also not uniformly available. Google lists developer access through the Gemini API and AI Studio, with other products rolling out separately; voice replication in AI Studio is unavailable in the EEA, UK, Switzerland, India, Illinois and Texas. For teams considering the models, the practical questions are whether directed performances remain consistent across long projects, how well outputs work in the target language, and what consent and watermark controls look like in actual use. The announcement establishes a new set of controls and access points, not independent proof of production quality.