Google has officially launched a new generation of artificial intelligence voice generation models, including “Gemini 3.8 Flash TTS” and “Flash Lite,” describing them as its most expressive and emotion-conveying audio tools to date, while offering original voice creation capabilities via text prompts in more than 100 global languages and dialects.
Article index:
- Launch of Flash and Flash Lite models via APIs
- Custom voice creation and expansion to 2,000 ready options
- Voice tone cloning feature, consent controls, and privacy protection
- Dramatic direction, non-verbal effects, and podcast scenes
- Topping international performance tests and global quality benchmarks
- Platforms, certified partners, and digital content dubbing
- Frequently asked questions
Launch of Flash and Flash Lite models via APIs
Google announced on Wednesday the rollout of two advanced models for converting written text into hyper-natural speech: “Gemini 3.8 Flash TTS” and “Gemini 3.8 Flash Lite TTS,” describing the new technology as the most important and expressive leap in the history of voice generation tools developed by the company.
The immediate rollout of the two models began via Gemini APIs and the Google AI Studio platform. The Flash model is designed to provide deep artistic and dramatic direction capabilities for creators, while the Flash Lite model is dedicated to large-scale commercial applications that require processing massive volumes of text at benchmark response speeds and a low financial cost.
Custom voice creation and expansion to 2,000 ready options
The new ecosystem allows developers and content creators to generate original and unique human voices entirely from scratch by writing natural language prompts, in which the user specifies the dramatic role, required dialect, and specific vocal traits such as a rough tone, emotional warmth, or news-anchor dignity.
This feature supports over 100 languages and local dialects worldwide. Google has expanded its official voice library available to businesses from 30 initial voices to more than 2,000 professional voice options ready for immediate production, covering precise geographic and cultural variations such as Mexican Spanish, Canadian French, and Scottish English.
Voice tone cloning feature, consent controls, and privacy protection
Google equipped the model with an impressive engineering feature that allows cloning and reconstructing the vocal fingerprint of any speaker with astonishing accuracy based on a recorded voice sample lasting no more than 30 seconds, provided the user holds the legal rights to that vocal fingerprint.
The company imposed strict security restrictions, as the system requires explicit audio recording of verbal consent from the original voice owner. All generated clips incorporate inaudible digital watermarks via SynthID technology and C2PA cryptographic certificates, while the feature’s activation is banned in the US states of Illinois and Texas, European Union countries, the United Kingdom, Switzerland, and India to strictly comply with biometric data protection laws.
Dramatic direction, non-verbal effects, and podcast scenes
Both models support line-by-line directorial guidance, allowing the tone and speed of each individual sentence to be adjusted, and generating prolonged texts spanning hours of continuous audio broadcasting without any disruption in delivery pace, alongside a two-person original conversational scenes feature that allows recording podcast episodes and dramatic performances with tight, alternating dialogue.
Vital non-verbal effects can also be inserted within written dialogue texts, including automatic laughter, sighs, breathing tones, and interactive listening affirmations during the other party’s speech, giving the generated voices a living human character that breaks the usual mechanical monotony.
Topping international performance tests and global quality benchmarks
The new technology earned exceptional praise in independent global benchmarking evaluations. The Gemini 3.8 Flash TTS model claimed first place globally in the voice design test by Hume AI, scoring 71.4 points, and topped the accent simulation test with 60.8 points.
Both models ranked first and second in the overall digital audio quality index, and swept blind human preference surveys conducted via the Voice Arena platform, achieving the highest ratings in natural pronunciation quality and rhythm harmony across phonetically highly complex languages like Japanese, Brazilian Portuguese, Vietnamese, and Hindi.
Platforms, certified partners, and digital content dubbing
Major cloud software development platforms such as Agora, LiveKit, and Vercel have begun supporting Gemini audio models via APIs, while leading global companies such as Figma, HeyGen, and Wondercraft have rushed to integrate the models into their software to provide instant dubbing services, media translation, and multilingual customer service agents.
Google confirmed that dedicated enterprise access for major corporations via the Gemini Enterprise bundle will be available soon, paving the way for a comprehensive transformation in the radio broadcasting and advertising industries, as well as the development of smart personal assistants in cars and mobile phones.
Frequently asked questions
Question: What are the two new models launched by Google for voice generation?
Answer: Gemini 3.8 Flash TTS was launched for advanced creative direction, and Flash Lite for intensive, low-cost applications.
Question: How many languages and voice options does Google’s new library support?
Answer: The models support over 100 languages and dialects, and their official library includes more than 2,000 professional voices ready for production.
Question: What is the audio duration required to accurately clone a speaker’s tone?
Answer: The system requires a recorded voice sample lasting only 30 seconds, along with explicit verbal consent to ensure security.
