Google has expanded its AI audio portfolio with two new text-to-speech systems: Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. The models are designed for different voice-generation workloads, ranging from creative audio production to large-scale automated speech services.

The release gives developers more options for creating synthetic voices from written prompts. Google is positioning the full Flash model toward applications that require expressive and controllable speech, while the Lite version is aimed at environments where organizations need to generate large amounts of audio efficiently.

Gemini 3.8 Flash TTS Targets Creative Voice Production

The standard Gemini 3.8 Flash TTS model is intended for projects that require greater control over how generated speech sounds. Potential applications include video games, interactive experiences, character dialogue, audiobooks, podcasts, and other forms of long-duration narration.

Instead of limiting developers to a small collection of predefined speakers, Google says its latest systems provide access to more than 2,000 ready-made voice profiles.

These profiles cover a wide range of linguistic and regional characteristics. Supported examples include Quebec French, Scots English, and Mexican Spanish, with voice generation available across more than 100 languages.

Google is also preparing a voice remixing capability. The feature is expected to let creators modify characteristics such as vocal tone, pitch, speaking speed, and accent through natural-language instructions.

Flash-Lite Focuses on High-Volume Audio Tasks

Gemini 3.8 Flash-Lite TTS has a different purpose. Google designed the smaller system for workloads where speech needs to be produced repeatedly at scale.

Possible use cases include automated dubbing, multilingual content conversion, customer support agents, and translation services.

This approach could be useful for companies processing large volumes of spoken content, particularly when efficiency and infrastructure costs are important considerations.

The two models therefore serve different production requirements rather than simply offering two versions of the same voice generator.

New Models Expand Google’s Audio AI Portfolio

The Gemini 3.8 Flash TTS launch follows Google’s earlier work in AI-powered speech and audio processing.

The company’s broader lineup has included systems such as Gemini 3.5 Live Translate, Gemini 3.5 Transcribe, Gemini 3.8 Live, and Gemini 3.8 Live Extended Thinking.

The latest release further expands Google’s focus on conversational and generative audio technology.

The move from a limited collection of fixed voices to thousands of selectable profiles also gives developers more flexibility when designing voices for different audiences, characters, regions, and applications.

Benchmark Results Highlight Voice Quality

Third-party testing has provided additional data about the performance of Gemini 3.8 Flash TTS.

On the Hume AI Voice Design Benchmark, the larger model achieved an overall score of 71.4, including a 60.8 result for accent modelling.

The model also received strong results in Hume AI’s overall quality evaluation. Gemini 3.8 Flash TTS reached the highest position in that assessment, while Flash-Lite TTS placed second.

These evaluations provide an external measurement of the models’ speech-generation capabilities, although benchmark scores do not necessarily represent performance across every real-world application.

Human Testing Examines Multilingual Speech

Google’s announcement also points to double-blind human evaluations conducted through Voice Arena.

The testing included several languages, such as Japanese, Brazilian Portuguese, Vietnamese, Modern Standard Arabic, Mexican Spanish, and Hindi.

These evaluations focused on listener preferences and provide additional information about how the generated voices perform outside English-language scenarios.

Multilingual performance is becoming increasingly important for AI voice systems as businesses use synthetic speech for international customer support, localization, education, entertainment, and media production.

Longer Dialogues Can Use Multiple Speakers

Another feature of the new models is support for multi-speaker conversations.

Developers can create scripts involving two speakers while allowing the generated voices to maintain separate identities throughout the exchange. This can be useful for podcasts, dialogue-heavy games, training material, virtual assistants, and other conversational applications.

The system is also designed for longer generation sessions. Maintaining a consistent voice over extended recordings can be particularly important for audiobooks, serialized content, and lengthy narration projects.

Developers Can Add Natural Speech Reactions

The models support additional control through text-based instructions for non-verbal sounds and conversational reactions.

For example, creators can place markers representing actions such as laughter, sighing, or gasping within a script. Short conversational responses can also be included to make exchanges sound less rigid.

These controls can help writers create dialogue with pauses, reactions, and changes in conversational rhythm rather than producing speech that follows only plain written sentences.

Google Adds Identity Verification for Voice Customization

Voice customization also introduces security concerns, particularly when technology can reproduce characteristics associated with a real person.

Google says its voice-cloning process requires identity verification. A custom voice request uses a 30-second reference recording together with a separate consent recording spoken by the person whose voice is being used.

The company checks the relationship between the recordings before allowing the custom voice to be processed.

This requirement is designed to reduce unauthorized attempts to reproduce someone’s voice and adds an additional verification step to the creation process.

AI-Generated Audio Includes Provenance Technology

Google is also applying technology designed to help identify synthetic audio.

Generated files contain an inaudible SynthID watermark, while C2PA provenance metadata is included with exported audio.

These mechanisms can help downstream systems determine whether an audio file was produced synthetically and provide information about its origin.

Provenance features are becoming increasingly relevant as generated voices become more difficult to distinguish from recordings made by human speakers.

Gemini TTS Models Reach Developer Platforms

Developers can access Gemini 3.8 Flash TTS and Flash-Lite TTS through Google AI Studio and the Gemini API.

The models can also be connected with development ecosystems involving platforms such as Agora, LiveKit, Pipecat, and Vercel.

Several companies are already exploring the technology for applications involving media generation, localization, translation, and customer communication. Early integrations mentioned by Google include Figma, HeyGen, Linguana, Wondercraft, 99.co, and Ollang.

These integrations demonstrate how synthetic voice technology can be incorporated into existing software rather than requiring companies to build separate audio-generation systems from scratch.

Gemini and Google Vids Get New Voice Capabilities

Google is making the technology available across several of its own products.

Users can access Flash TTS through Gemini Notebook, while Google Vids uses the Flash-Lite system for its voice-generation capabilities.

Google also plans to provide Gemini Enterprise customers with administrative API access as the rollout expands.

What the Gemini 3.8 Flash TTS Launch Means

The introduction of Gemini 3.8 Flash TTS and Flash-Lite TTS gives developers two speech-generation options aimed at different workloads.

The full Flash model emphasizes expressive voices, multilingual control, speaker consistency, and creative production. Flash-Lite is designed around scalable audio generation for applications such as dubbing, translation, and automated communication.

With thousands of voice profiles, support for more than 100 languages, conversational controls, identity verification, and audio provenance features, Google’s latest TTS release combines voice creation with safeguards intended for commercial and developer use.

As AI-generated speech continues to move into entertainment, software, media, and customer service, these capabilities could make synthetic voice production more flexible and easier to integrate into existing workflows.