Google DeepMind launched two new text-to-speech models on September 23, 2026, that fundamentally shift how synthetic voices are created: instead of selecting from a catalog, developers now describe a character in plain language and the AI generates a matching voice. The release marks Google's third audio model launch in under 30 days, signaling that voice AI has become a priority battleground in the broader AI market.

What Makes This Different From Traditional Text-to-Speech?

The headline capability is voice design from natural language prompts. Developers can describe a voice's role, accent, personality, and emotional tone in plain English, and Gemini 3.8 Flash TTS generates an entirely new voice that matches the brief. This is a meaningful departure from how most text-to-speech systems work today, which typically offer a fixed menu of pre-recorded or pre-trained voices.

The second major feature is line-by-line performance direction. Instead of applying one tone across an entire script, developers can attach acting cues, pacing instructions, dialect shifts, and backchanneling (small verbal acknowledgments like "mm-hmm" that make dialogue sound human) to individual lines. This level of granular control treats the AI voice like an actor who can shift energy and emotion mid-scene, rather than a screen reader that maintains consistency throughout.

"Built for deep creative direction and character design," capable of designing entirely new voices from natural-language prompts across more than 100 languages and dialects," stated Google DeepMind in its official announcement.

Google DeepMind, Official Announcement

The emphasis on dialects, not just top-level languages, is a distinguishing detail. Most competing text-to-speech catalogs list languages but treat regional variation (Mexican Spanish versus European Spanish, for instance) as a secondary feature rather than a core design axis. Google's approach plays directly into localization workflows for global publishers, whether that is a game studio dubbing dialogue for a dozen regional markets or a media company producing region-specific ad reads from one script.

How Do the Two Models Work Together?

Google released two distinct models designed to work at different stages of a production pipeline. Gemini 3.8 Flash TTS is the creative-direction model, optimized for crafting a specific character voice that will then be reused across a project. Gemini 3.8 Flash-Lite TTS is the deployment model, built for high-volume, cost-sensitive generation once a voice or style has already been settled on.

• Flash TTS Use Case: Creative direction and character design, where developers spend time iterating on a unique voice that matches their creative vision before locking it in for production.
• Flash-Lite TTS Use Case: High-volume generation and voice agents, powering thousands of customer service calls or narrating video content at scale through Google Vids.
• Language Coverage: Both models support more than 100 languages and dialects, enabling global content creators to work from a single system rather than managing separate vendor contracts per region.
• Voice Library: Flash TTS lets developers design new voices from text prompts, while Flash-Lite TTS draws from either developer-created styles or Google's production-ready library of over 2,000 voices.

How to Implement Voice Design in Your Audio Workflow

• Define Your Character in Plain Language: Write a natural-language description of the voice you want, including accent, age range, emotional tone, and personality traits. The model will generate a voice matching your brief without requiring voice talent or stock library selection.
• Add Line-by-Line Performance Direction: Attach acting cues, pacing instructions, and dialect shifts to individual lines of your script. This allows a single voice to sound different in different emotional beats, eliminating the need for manual audio editing or multiple generation passes.
• Leverage Regional Dialect Support: Use the 100+ language and dialect options to produce region-specific content from a single script. This simplifies localization workflows for game studios, audiobook producers, and media companies working across multiple markets.
• Choose the Right Model for Your Stage: Use Flash TTS for creative iteration and character design, then switch to Flash-Lite TTS for high-volume production once your voice is finalized. This two-stage approach balances creative control with cost efficiency at scale.

Both models began rolling out through the Gemini API and Google AI Studio on launch day, with Flash TTS also available in Gemini Notebook and Flash-Lite TTS integrated into Google Vids. Gemini Enterprise API access was listed as coming soon rather than immediately available.

Where Does This Fit in the Competitive Landscape?

The announcement lands at a moment when voice AI has become one of the more contested corners of the AI market. ElevenLabs built a business on expressive voice cloning and dubbing, while OpenAI, Amazon, and Microsoft all ship competing speech stacks. Google's strategic shift is to stop treating text-to-speech as a narration utility and start treating it as a creative tool with actors, accents, and stage directions built into the prompt.

Logan Kilpatrick, who leads developer relations for Google AI, noted that the model claimed the top spot on Hume AI's voice benchmarks, which focus specifically on how natural and emotionally convincing synthetic voices sound to listeners, not just intelligibility. This benchmark claim is significant because it suggests Google's approach to performance direction and character design may produce voices that feel more human and emotionally authentic than systems optimized purely for clarity.

The pace of Google's releases also matters strategically. Three separate audio model launches inside a single month is an aggressive cadence even by Gemini's recent standards, signaling that Google is treating voice as a priority battleground rather than a side feature. For studios and narrative-heavy publishers who have historically avoided text-to-speech because it could not act, this level of granular control raises the bar for what counts as "good enough" synthetic voice technology.

Google has not published a full, independently verified rate card as of the launch date, so any pricing figures circulating on developer forums should be treated as provisional until confirmed on Google's official documentation.