Google Launches Gemini 3.8 Flash TTS: Voice Cloning and Text-Described Voices for $9-18 per Million Audio Tokens

Google has released two new text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, that let users design synthetic voices from text descriptions rather than choosing only from a fixed library. Both models support more than 100 languages and are rolling out now through the Gemini API and Google AI Studio.

Text-described voices and cloning

With Flash TTS, a text prompt can define a voice's role, accent, and vocal traits, according to Google. Users who prefer a preset can pick from a library of more than 2,000 built-in voices, including regional variants like Mexican Spanish, Quebec French, and Scottish English.

A voice cloning feature builds a voice profile from a 30-second audio sample. Google requires the person being cloned to record a spoken consent statement that matches the voice in the sample. Every generated clip carries an inaudible SynthID watermark intended to help detect AI-generated speech, according to the company.

Google also previewed a feature called "Voice Remixing," which will let users adjust the timbre, pitch, tempo, and accent of library voices. It is not yet available.

Stage directions and dialogue control

Both models accept per-line stage directions or can interpret script cues automatically. Google says the models can generate hours of audio with minimal "speaker drift" — meaning the voice stays consistent over long outputs. A two-voice mode generates dialogue from a single script while keeping each speaker distinct, and users can script nonverbal sounds like laughter, sighs, and filler words such as "mhm" to control timing and pacing.

Google claims its models lead in most categories of Hume AI's text-to-speech benchmark, though the company has not published specific scores alongside the release.

Two models, two use cases

Gemini 3.8 Flash TTS targets creative production — podcasts, audiobooks, and game character voices. Gemini 3.8 Flash-Lite TTS is positioned for lower-cost speech generation at scale, aimed at dubbing, audio content, and voice agents.

Pricing is billed in US dollars per million tokens, with text tokens for input and audio tokens for output. Through the end of 2026, Flash TTS costs $0.50 per 1M input tokens and $9.00 per 1M output tokens; Flash-Lite TTS costs $0.50 per 1M input tokens and $6.00 per 1M output tokens. Both rates roughly double starting January 1, 2027, rising to $1.00/$18.00 for Flash TTS and $1.00/$12.00 for Flash-Lite TTS.

Google says one second of generated audio equals 25 audio tokens, putting an hour of output at 90,000 tokens. That works out to $0.81 per hour of audio with Flash TTS and $0.54 per hour with Flash-Lite TTS under 2026 pricing, rising to $1.62 and $1.08 respectively in 2027.

Flash TTS is available in Gemini Notebook, and Flash-Lite TTS is available in Google Vids. Gemini Enterprise API access is expected to follow for both models. Developer platforms including Agora, LiveKit, Pipecat, and Vercel already support integration through the Gemini API. Google has not yet listed regional endpoints for the new models, though earlier TTS models offered EU data processing. Data from free-tier usage is used to improve Google's products; paid-tier data is not, according to the company.

What this means

The split between a premium creative model and a low-cost scaled model signals Google is targeting two distinct buyer types: studios producing narrative audio and enterprises running voice agents or dubbing pipelines at volume. Text-described voice design and 30-second cloning lower the barrier to custom voice production, but the consent-recording requirement for cloning shows Google is trying to preempt misuse concerns before regulators do. The near-doubling of prices in 2027 suggests introductory rates meant to drive adoption before Google normalizes margins on the audio output side, which is priced far higher than text input.