Google releases Gemini 3.8 text-to-speech models with voice design and replication

Archive entry: this covers an event from September 22, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On September 22, 2026, Google made gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts generally available, together with a /v1beta/voices endpoint, as recorded in the Gemini API changelog. The text-to-speech models supported voice design and replication, and their output carried provenance marks.

What changed with Gemini 3.8 TTS

• gemini-3.8-flash-lite-tts replaced gemini-3.1-flash-tts-preview.
• A voice could be replicated from a 30-second sample plus a verbal consent recording.
• Output carried SynthID watermarks and C2PA metadata.
• The models supported 100+ languages, expressive tags such as one for laughter, and two-speaker dialogue staging.
• The changelog cited a library of 150+ voices, while Google's blog post on Gemini 3.8 text-to-speech, dated September 23, 2026, cited 2,000+ production-ready voices, up from 30 original voices.
• Google cited Hume's benchmarks, where the model ranked #1 in Voice Design (71.4) with an accent-modeling score of 60.8 and held the #1 and #2 positions for Overall Quality.

What it means for people who build with AI

Developers got production text-to-speech with custom and cloned voices, plus built-in consent and provenance controls. Voice replication paired a 30-second sample with a verbal consent recording, and generated speech carried SynthID watermarks and C2PA metadata.

Availability and next steps

The models were available in the Google AI Studio playground and for everyone in Gemini Notebook and Google Vids. Voice remixing was announced as a feature for a later release.

Sources

• Gemini API changelog: September 22, 2026 (Google AI Studio (Gemini API changelog), September 22, 2026, official)
• Gemini 3.8 text-to-speech says hello (Google (The Keyword), September 23, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.