Gemini 3.8 text-to-speech says hello

Google has just made a significant stride in the world of artificial intelligence, unveiling its latest generation of voice AI models. This week, we're not just hearing about incremental updates; we're witnessing a genuine breakthrough in how AI can generate and interact with human speech. The spotlight is on the new Gemini 3.8 Flash TTS and Flash-Lite TTS models, which promise to transform text into highly expressive, customizable, and natural-sounding audio. These models, announced on September 23, 2026, are set to redefine the landscape of AI-powered voice generation across various applications.

This release is part of a broader "Gemini 3.8" family of advancements, which also includes the Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking models, launched just a week prior on September 15, 2026. While the Live models focus on real-time, bidirectional speech-to-speech interactions for conversational agents, the Flash TTS models specifically elevate the art of generating speech from text. This comprehensive suite of tools underscores Google's commitment to pushing the boundaries of human-like AI communication.

The Dawn of Expressive AI Voices: Introducing Gemini 3.8 Flash TTS and Flash-Lite TTS

The core of this breakthrough lies in the new Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. These are not just improved versions of existing text-to-speech systems; they represent a leap forward in expressive audio generation. Google, the developer behind these models, states that they are their most expressive yet. The models began rolling out on September 23, 2026, and are accessible through the Gemini API and Google AI Studio, among other Google products and enterprise surfaces.

The distinction between the two Flash TTS models is clear: Gemini 3.8 Flash TTS is tailored for creative direction and intricate character design, making it ideal for fields like gaming, immersive audiobooks, podcasts, and interactive media. On the other hand, Gemini 3.8 Flash-Lite TTS is optimized for high-volume, cost-efficient applications such as dubbing, general audio content creation, and voice agents that require precise control over tone, pacing, and expressive nuances. This strategic split allows developers and creators to choose the model that best fits their specific needs, balancing quality and scale.

Unpacking the Breakthrough: Key Capabilities

The Gemini 3.8 Flash TTS models bring a host of advanced features that truly distinguish them from previous generations and current competitors:

Voice Customization and Design

One of the most exciting capabilities is the ability to design bespoke voices from scratch using natural language prompts. Users can describe the role, accent, and specific voice characteristics, and the AI will generate a unique voice. This functionality spans over 100 languages and dialects, allowing for incredibly nuanced and culturally specific voice creation. Google has also expanded its library to over 2,000 production-ready voices, including regional variations like Mexican Spanish, Quebec French, and Scots English. This vast selection provides creators with an unprecedented palette for their audio projects.

Voice Replication

For those needing to maintain a consistent vocal identity, Gemini 3.8 Flash TTS offers robust voice replication. It can recreate a stable vocal profile from as little as a 30-second audio sample. Google emphasizes an ethical approach here, requiring users to provide a verbal consent recording from the voice owner that matches the reference speaker before a voice can be created. This crucial step helps protect voice talent and ensures responsible use of the technology.

Expressive Control

The new models provide granular control over vocal delivery, transforming voice generation from a static process into a dynamic creative studio. Creators can now direct vocal performances line by line, incorporating "stage directions" directly into their scripts. This includes embedding emotion and tone tags into prompts, specifying dramatic pause markers for precise timing, and managing multi-speaker dialogue within a single script. These features make it possible to generate long-form content and complex screenplay dialogue with natural turn-taking and emotional depth, previously challenging for AI.

Multilingual Mastery

Global reach is a cornerstone of Gemini 3.8. With support for over 100 languages, the models empower creators to build high-quality, multilingual voice experiences worldwide. This includes not just language support but also the ability to handle regional accents and dialects. For the Gemini 3.8 Live models, there's even the capability to automatically detect and switch between 97 supported languages mid-conversation, making real-time global interactions incredibly fluid.

Integrated Safety and Ethics

Google is prioritizing responsible AI development. Every audio clip generated by Gemini Audio models, including the new Flash TTS variants, is watermarked with SynthID. This digital watermark is designed to help detect AI-generated content and combat misinformation. Additionally, replicated voices carry C2PA content credentials, further enhancing transparency and traceability. Google's comprehensive safety assessments have indicated that these models do not introduce new critical capabilities that would raise frontier safety concerns compared to previous versions.

Beyond Text-to-Speech: The Gemini 3.8 Live Ecosystem

It's important to view the Gemini 3.8 Flash TTS models within the broader context of the Gemini 3.8 family. The Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking models, released on September 15, 2026, represent a parallel breakthrough in real-time conversational AI. These are native speech-to-speech models, meaning they directly convert audio input to audio output, bypassing traditional cascaded Speech-to-Text (STT), Large Language Model (LLM), and Text-to-Speech (TTS) pipelines.

The defining feature of the Live models is their ability to maintain a fluid conversation even while reasoning and executing tool calls in the background. They support multimodal input, processing text, images, video, audio, and PDFs, and can output both text and audio. Gemini 3.8 Live is built for efficient, high-volume voice agent deployments, while Extended Thinking is designed for complex, multi-step workflows, capable of reasoning and speaking simultaneously to provide spoken progress updates during longer tasks. These advancements are crucial for building highly intelligent and responsive voice agents, further enhancing the overall Gemini 3.8 voice AI capabilities.

Performance and Benchmarks

The new Gemini 3.8 models aren't just rich in features; they also demonstrate leading performance in independent evaluations. Gemini 3.8 Flash TTS has secured the #1 overall spot on Hume AI's Voice Design Benchmark with a score of 71.4, and it also leads in accent modeling with 60.8. Both Gemini 3.8 Flash TTS and Flash-Lite TTS have earned the #1 and #2 spots, respectively, on Hume AI's Overall Quality Index, indicating superior expressive performance without sacrificing reliability. In blind human preference evaluations on Voice Arena, these models achieved top positions across key global languages, including Japanese, Brazilian Portuguese, Vietnamese, Modern Standard Arabic, Mexican Spanish, and Hindi.

For the speech-to-speech domain, Gemini 3.8 Live Extended Thinking has captured the #1 overall spot on Artificial Analysis' Speech to Speech Quality Index (82.6), showcasing its intelligence and enterprise-grade task completion capabilities. It also leads in agentic task completion benchmarks like τ-Voice. These strong benchmark results underscore the technical prowess and real-world applicability of the Gemini 3.8 voice AI family.

Accessibility and Availability

Google has made these advanced models broadly accessible to developers and enterprises. The Gemini 3.8 Flash TTS and Flash-Lite TTS models are rolling out in the Gemini API and Google AI Studio. The Gemini 3.8 Live and Extended Thinking models are also available through the Gemini API, Google AI Studio, and are being integrated into Google Workspace, Search, and the Gemini app, making these powerful capabilities available across Google's ecosystem.

Pricing Structure

Understanding the pricing for these advanced models is crucial for developers and businesses. Unlike older Text-to-Speech models that often charge per character, Gemini-TTS models (including Gemini 2.5 Flash TTS, Gemini 3.1 Flash TTS, and the new 3.8 variants) utilize a different billing approach. Google charges for both text input tokens and generated audio tokens. The audio charge typically accounts for the majority of the cost for standard narration.

For instance, current rates indicate that Gemini 2.5 Flash TTS costs approximately $0.015 per minute of generated audio, plus a small text-input charge. Gemini 3.1 Flash TTS Preview and Gemini 2.5 Pro TTS are around $0.03 per generated minute. This translates to approximately $0.91 or $1.81 per hour of output, respectively, under ordinary pacing assumptions. While specific pricing for Gemini 3.8 Flash TTS and Flash-Lite TTS would align with this token-based model, the Gemini 3.8 Live model is priced at approximately $0.84 per hour of input audio, and Gemini 3.8 Live Extended Thinking at around $3.50 per hour of input audio. This token-based approach aims to provide a more granular and potentially more cost-effective solution for advanced voice AI applications compared to character-based billing for older models.

Industry Implications: What This Means for Voice AI

The release of Gemini 3.8 Flash TTS and the broader Gemini 3.8 voice AI family marks a pivotal moment for several industries:

• Content Creation: Audiobooks, podcasts, video game development, and film dubbing will see a dramatic improvement in quality and efficiency. Creators can achieve nuanced performances without the extensive time and cost typically associated with human voice talent.
• Customer Service and Virtual Assistants: The Gemini 3.8 Live models, with their real-time, multimodal, and multi-language capabilities, will enable more natural, intelligent, and helpful conversational AI agents. This could lead to genuinely fluid interactions that feel less robotic and more human.
• Accessibility: High-quality, expressive text-to-speech can significantly enhance accessibility for individuals with visual impairments or reading difficulties, providing more engaging and understandable auditory content.
• Education: Personalized learning experiences can be enriched with AI voices that adapt their tone and style to better engage students or explain complex concepts.
• Global Communication: With robust multilingual support and mid-conversation language switching, these models can break down language barriers in real-time interactions and content localization.

This breakthrough isn't just about making AI voices sound more human; it's about giving creators and developers unprecedented control to craft specific vocal identities and experiences. The integration of advanced safety measures also sets a new standard for ethical AI deployment, fostering trust and transparency in AI-generated content.

The Road Ahead for Google's Voice AI

Google's Gemini 3.8 voice AI models represent a significant step towards a future where human-AI interactions are seamless, natural, and incredibly rich. By offering both highly expressive text-to-speech and sophisticated real-time speech-to-speech capabilities, Google is laying the groundwork for a new generation of intelligent agents and immersive audio experiences. The emphasis on creative control, multilingual support, and ethical safeguards positions Gemini 3.8 not just as a technological marvel but as a responsible advancement in the evolving landscape of artificial intelligence. As these models become more widely adopted, we can expect to see innovative applications that were previously unimaginable, further blurring the lines between human and synthetic communication.

Frequently Asked Questions

What is Gemini 3.8 Flash TTS?

Gemini 3.8 Flash TTS is Google's latest text-to-speech AI model, released on September 23, 2026, designed for highly expressive and customizable voice generation from text. It excels in creative applications like audiobooks and games, offering features like natural language voice design and voice replication.

How does Gemini 3.8 Flash TTS differ from Flash-Lite TTS?

Gemini 3.8 Flash TTS is optimized for deep creative direction and character design, while Gemini 3.8 Flash-Lite TTS is built for high-volume, cost-efficient use, such as dubbing and general audio content creation, offering fine-grained control over tone and pacing.

Can Gemini 3.8 Flash TTS clone voices?

Yes, Gemini 3.8 Flash TTS can recreate a consistent vocal profile from a 30-second audio sample. Google requires verbal consent from the voice owner, and the system verifies this consent to ensure ethical use.

What are Gemini 3.8 Live and Extended Thinking models?

Gemini 3.8 Live and Extended Thinking are complementary speech-to-speech AI models, released on September 15, 2026, focused on real-time conversational agents. They handle multimodal input (audio, text, images, video) and can execute tasks in the background without interrupting dialogue, making AI interactions more fluid and intelligent.