TechPrufer
Get started
DeepMind
DeepMind logoDeepMind · Feature trail

Live Speech-to-Speech Translation

3 posts · Dec 2025 – Sep 2026

What this trail covers

Gemini models are enhanced for live voice interactions, including improved speech-to-text and speech-to-speech translation. Gemini 3.8 Live and 3.8 Live Extended Thinking offer advancements in near real-time reasoning for voice agents, with improved conversational intelligence, fluid dialogue, visual grounding, and multi-step reasoning. These models support over 97 languages mid-conversation, execute tools and API calls in the background, and provide live progress narration for complex tasks. The Live Avatar feature adds near real-time visual presence to these models, enabling dynamic visual personas with precise lip-syncing, natural expressions, and fluid turn-taking. This allows for richer, more accessible enterprise experiences, asynchronous tool execution with continuous presence, and native multilingual speech-to-speech synchronization across 97 languages. Organizations can also customize avatars to fit brand needs, with safeguards like SynthID watermarking for transparency.

All DeepMind trails
  1. Introducing Gemini 3.8 Live with Live AvatarIntroduces Gemini 3.8 Live with Live Avatar, adding near real-time visual presence to existing live dialogue models. This feature enables dynamic visual personas with precise lip-syncing, natural expressions, and fluid turn-taking. It supports asynchronous tool execution with continuous presence, allowing background tool calls while maintaining active dialogue. The feature offers native multilingual speech-to-speech synchronization across 97 languages without degrading video fidelity. It also allows for custom avatar generation from reference images to match brand needs. SynthID watermarking is integrated for transparency and to minimize misinformation.
  2. Introducing Gemini 3.8 Live and 3.8 Live Extended ThinkingIntroduces Gemini 3.8 Live and 3.8 Live Extended Thinking models. Gemini 3.8 Live focuses on scale and cost efficiency with conversational intelligence, fluid dialogue, and visual grounding, achieving high user preference and cost-effectiveness. Gemini 3.8 Live Extended Thinking is designed for high-complexity tasks with increased intelligence and multi-step reasoning, leading in agentic task completion benchmarks and providing strong reasoning capabilities. Both models process visual inputs in near real-time, support automatic language detection and transition across 97 languages, and execute tools/API calls in the background. Extended Thinking provides simultaneous reasoning and speaking with early verbal cues and live progress narration. The Gemini Live API is updated to support these models, with integrations from platforms like Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents. SynthID watermarking is applied to all generated audio.
  3. Fluid, natural voice translation with Gemini 3.5 Live TranslateIntroduces Gemini 3.5 Live Translate, a new audio model for live speech-to-speech translation. It automatically detects 70+ languages and generates natural-sounding translated speech that preserves intonation, pacing, and pitch. The model balances context for quality with immediate translation for synchronization, delivering fluid audio with minimal delay. It is available via Gemini Live API, Google AI Studio, Google Meet, and the Google Translate app. The post highlights its use in developer platforms like Agora, Fishjam, LiveKit, Pipecat, and Vision Agents, and mentions testing by Grab. It also details improvements in Google Meet's speech translation capabilities and a new 'listening mode' in the Google Translate app for Android. All generated audio is watermarked with SynthID.
  4. Gemini 3.1 Flash Live: Making audio AI more natural and reliableGemini 3.1 Flash Live is introduced as a high-quality audio and voice model for real-time dialogue, offering improved speed and natural rhythm for voice-first AI. It achieves 90.8% on ComplexFuncBench Audio for multi-step function calling and 36.1% on Scale AI's Audio MultiChallenge for complex instruction following and long-horizon reasoning. It also features improved tonal understanding for dynamic response adjustment based on user expressions and is inherently multilingual, supporting over 85 languages. All audio output is watermarked with SynthID.
  5. Improved Gemini audio models for powerful voice interactionsThis post details improvements to Gemini 2.5 Flash Native Audio for live voice agents, focusing on sharper function calling (71.5% on ComplexFuncBench Audio), robust instruction following (90% adherence), and smoother multi-turn conversations. It also introduces live speech-to-speech translation with continuous listening and two-way conversation capabilities, supporting over 70 languages and preserving speaker intonation, pacing, and pitch. The translation feature is rolling out in the Google Translate app and will be integrated into the Gemini API in 2026.