Gemini Audio
Speech generation
Create rich custom voices, replicate reference audio, and orchestrate two-speaker dialogue with control over style, pitch and performance.
Models
Generate unique life-like vocal identities — or tap into over a thousand production-ready voices in our pre-crafted library.
3.8 Flash TTS
Best for generative voice design and two-speaker dialogue generation. Keeps intonation, inflection, and pace steady from sentence-to-sentence, as you direct with precision.
3.8 Flash-Lite TTS
Best for everyday audio generation, like video dubbing or podcast creation. Balances efficiency with expressiveness to power large content libraries and high-traffic apps.
Capabilities
Generate speech with precise control to create emotive narratives – or create entire conversations featuring multiple speakers.
Generative voice design
Create a rich cast of voices using simple natural language prompts. Describe any persona — from warm documentary narrator to eccentric fantasy character — or fine-tune traits like accent, age, pitch, and texture on demand.*
*Gemini 3.8 Flash TTS only.
Scripted dialogue control
Direct single-narrator and multi-character scenes with clear voice distinction. Separate dialogue from acting notes for control over emotion, tone, and volume.
Voice library and regional presets
Choose from thousands of production-ready voices in different global accents, dialects and languages.
Voice replication
Turn seconds of reference audio into a high-fidelity digital voice. Our secure, consent-based workflow lets you create accurate voice copies for dubbing, narration, and scaling voice talent.
Granular expressive control
Use intuitive audio tags to command style, pace, and delivery with unprecedented precision.
Showcase
Slide 1 of 11
“Gemini 3.8 Flash TTS brings AI voice conversations much closer to feeling genuinely human. The expressiveness, consistency and support for local languages and accents, from Singlish to Indonesian, have been stellar. We’re building furiously with it at 99.co for real-time AI phone conversations.”
Darius Cheung
CEO and Co-founder, 99.co
“A great voice agent needs to do more than speak clearly. It needs to sound right for the person, the language, and the moment. This new generation of text to speech gives developers more control over all three.”
Mason Adams
Developer Evangelist, Agora
“Gemini's new text-to-speech model lets you craft natural voices in Figma Weave with full creative control of the emotion, tone, and delivery. It turns sound design into another creative layer on the canvas, helping teams direct a full sensory story in one place.”
Jonathan Gur-Zeev
Director of Product, Figma Weave
“Gemini TTS' new Voice Design featured gave HeyGen users another great option to fully control and define the unique voice for their content, and more importantly leads to more consistent voices. Our customers have created over 20 million unique custom voices to date, and we are excited to incorporate the latest Gemini TTS capabilities into our platform.”
Bin Liu
VP of Engineering, HeyGen
“We chose Gemini 3.8 Flash TTS for Katsuyo as we found other models to be either overly dramaticized, or lacking in critical aspects of Japanese text-to-speech such as pitch accent and correct reading of kanji. It has the best accuracy and most natural voice for both emotionally charged content and everyday speech.”
Luke Pane
Developer, Katsuyo
“Gemini TTS is transforming how we create and deliver AI-powered audio. Its natural, expressive voices and multi-speaker capabilities enable us to produce high quality audio significantly faster, while maintaining consistency across different content formats. At our scale, this unlocks richer storytelling, faster content creation, and more engaging experiences for millions of users.”
Ritwik Baranwal
Associate Director - AI/ML, Kuku FM
“Linguana dubs and distributes YouTube creators' videos worldwide and produces realistic sponsored ad reads, using automated AI technology we've built in-house. Gemini TTS will take our quality even higher, with expressive, natural-sounding voices and the stability our pipeline relies on at scale. The result will be content that feels authentic, helping creators connect with audiences in every market.”
Oded Shafran
Co-Founder & CTO, linguana
“At Ollang, voice is a critical component of our AI Dubbing and Audio Description workflows, and the quality we're seeing with Gemini 3.8 Flash TTS opens a new level of what we can deliver for global media. It's a major step towards making multilingual voice AI sound genuinely human while preserving emotion, pacing, and intent.”
Aziz Ulak
CTO & Co-founder, Ollang
“At Spoken, The AI Audiobook Company™, our purpose-built Studio, trusted by authors and publishers, conducts deep analysis of written works to capture nuanced emotion, accent, and character direction. Gemini 3.8 Flash TTS translates that analysis into high-quality, human-like text-to-speech with expressive prosody and a large context window, making it a powerful addition to our premium Spoken Multi-Cast™ workflow.”
Phil Marshall
Founder and CEO, Spoken
"Voice is what makes a character believable. For our Simia app, keeping characters personable and consistent is crucial to our worldbuilding, and the latest Gemini TTS models bring a new level of craft to that work: more emotion in a scene, and voices steady enough to carry deeper storytelling."
Zina Rahman,
Co-Founder and CEO, Transforms AI
“Gemini 3.8 Flash-Lite TTS brings the voice consistency and creative control Wondercraft needs, making it a powerful fit for producing standout podcasts, audio ads, and voice-driven content at scale.”
Mei Ki Yiu
CTO, Wondercraft
Performance
Our speech generation models deliver impressively fast speech generation without compromising on vocal stability or expressive quality.
Text-to-Speech Quality Benchmark
Hume AI
On text-to-speech evals, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS secure the #1 and #2 spots on Hume AI’s Overall Quality Index, driving major improvements in long-form stability and dual-speaker screenplay control compared to Gemini 3.1 Flash TTS.
| Benchmark | Gemini 3.8 Flash TTS | Gemini 3.8 Flash-Lite TTS | Gemini 3.1 Flash TTS | Eleven Labs v3 | Eleven Labs v3 conversational | Cartesia Sonic 3.6 | OpenAI gpt-4o-mini-tts | Inworld TTS-2 |
|---|---|---|---|---|---|---|---|---|
| Overall Reliability x Expressiveness | 0.920 | 0.914 | 0.783 | 0.706 | 0.769 | 0.840 | 0.740 | 0.576 |
| Human-like variation Lower = flatter or wilder than a human across turns | 4.58 | 4.51 | 3.95 | 5.00 | 4.97 | 3.40 | 4.22 | 3.76 |
| Multispeaker | 4.14 | 4.10 | 3.60 | 3.85 | — | — | — | — |
| Style Tag Control Single Tag | 4.34 | 4.32 | 4.31 | 3.89 | 3.97 | 3.37 | — | 4.15 |
Text-to-Speech Voice Design Leaderboard
Hume AI
Gemini 3.8 Flash TTS’s voice design capabilities deliver frontier-level customization, securing the #1 overall spot on Hume AI’s Voice Design Benchmark (71.4) and leading the industry in accent modeling (60.8).
| Benchmark | Gemini 3.8 Flash TTS | ElevenLabs Voice Design v3 | Inworld Voice Design |
|---|---|---|---|
| Overall English | 71.4 | 70.8 | 69.8 |
| Multilingual | 3.82 | 3.65 | 3.57 |
| Accents | 60.8 | 45.4 | 35.8 |
| Voice Qualities Single Tag | 74.6 | 76.6 | 76.3 |
Text-to-Speech Leaderboard
Voice Arena
In blind human preference evaluations on Voice Arena, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS secure top positions amongst competitors in key global languages, including English, Japanese, Brazilian Portuguese, Vietnamese, and Modern Standard Arabic (MSA), Mexican Spanish and Hindi. With support for over 100 languages, these models empower creators, developers, and enterprises to build high-quality, multilingual voice experiences worldwide.
| Benchmark | Gemini 3.8 Flash TTS | Gemini 3.8 Flash-Lite TTS | Gemini 3.1 Flash TTS | ElevenLabs v3 | Cartesia Sonic 3.6 | OpenAI gpt-4o-mini-tts |
|---|---|---|---|---|---|---|
| English | 1061 | 1087 | 1051 | 986 | 1068 | 940 |
| Japanese | 1232 | 1152 | 1148 | 1048 | — | 975 |
| Brazilian Portuguese | 1104 | 1134 | 1094 | 1040 | 1080 | 946 |
| Vietnamese | 1135 | 1156 | 1099 | 1043 | — | 839 |
| Arabic MSA | 1204 | 1181 | 1135 | 1020 | — | 911 |
| Hindi | 1106 | 1076 | 1086 | 1052 | 1104 | 843 |
| Mexican Spanish | 1152 | 1146 | 1092 | 1015 | 1089 | 880 |
Model information
| Name | 3.8 Flash TTS | 3.8 Flash-Lite TTS |
|---|---|---|
| Status | General availability | General availability |
| Input |
|
|
| Output |
|
|
| Input tokens | 8K | 8K |
| Output tokens | 64K | 64K |
| Availability |
|
|
| Documentation | View developer docs | View developer docs |
| Model card | View model card | View model card |
Try Speech generation
Google AI Studio
The fastest path from prompt to production
Google Vids
AI-powered video creation for work
Gemini API
Get started building with cutting-edge AI models
Gemini Notebook
Your research and thinking partner