TechPrufer
Get started
DeepMind
DeepMind logoDeepMind · Feature trail

Expressive Text-to-Speech (TTS) Models

4 posts · Dec 2025 – Sep 2026

What this trail covers

Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS are introduced, enhancing voice generation with deep creative direction, character design, and high-volume, cost-efficient scaling. Gemini 3.8 Flash TTS allows for generative voice design from natural language prompts, voice replication from 30-second samples with consent verification and watermarking, and voice remixing. Both models offer granular control over line-by-line delivery, long-form generation, native two-speaker scene staging, and scripted vocal bursts. They achieve leading positions on voice design and quality benchmarks, supporting over 100 languages. The models are integrated into Google AI Studio, Gemini API, Gemini Notebook, and Google Vids, with enterprise offerings coming soon.

All DeepMind trails
  1. Gemini 3.8 text-to-speech says helloIntroduces Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, two new text-to-speech models. Gemini 3.8 Flash TTS enables generative voice design from natural language prompts, voice replication from 30-second audio samples with consent verification and SynthID watermarking, and voice remixing. Both models provide granular control over line-by-line delivery, long-form generation, native two-speaker scene staging, and scripted vocal bursts. The models achieve top rankings on Hume AI's Voice Design Benchmark and Overall Quality Index, and perform well in blind human preference evaluations on Voice Arena across multiple languages. They support over 100 languages and are integrated into Google AI Studio, Gemini API, Gemini Notebook, and Google Vids.
  2. Intelligent transcription with Gemini 3.5 TranscribeGemini 3.5 Transcribe is introduced as a precise speech-to-text model for intelligent voice interactions, improving upon Chirp 3. It offers real-time streaming and pre-recorded audio processing APIs. Key features include smart transcription (handling self-corrections, filler words, auto-formatting), function calling, improved Word Error Rate (WER) of 4.0% for streaming and 2.6% for non-streaming, custom vocabulary recognition, global language support for over 85 languages, and multi-speaker identification.
  3. Gemini 3.1 Flash TTS: the next generation of expressive AI speechIntroduces Gemini 3.1 Flash TTS, a new text-to-speech model with improved speech quality, controllability, and expressivity. Key technical contributions include: 1. Enhanced speech quality, achieving an Elo score of 1,211 on the Artificial Analysis TTS leaderboard. 2. Introduction of 'audio tags' for granular control over vocal style, pace, and delivery via natural language commands embedded in text input. 3. New developer experience features in Google AI Studio: scene direction for environmental context and dialogue instructions, speaker-level specificity with Audio Profiles and Director's Notes for pace, tone, and accent, and inline tags for mid-sentence expression changes. 4. Support for 70+ languages with advanced style, pacing, and accent control for localized experiences. 5. Audio watermarking with SynthID for reliable detection of AI-generated content.
  4. Gemini 3.1 Flash Live: Making audio AI more natural and reliableGemini 3.1 Flash Live is released as the highest-quality audio and voice model, offering speed and natural rhythm for voice-first AI. It excels in multi-step function calling (90.8% on ComplexFuncBench Audio) and long-horizon reasoning amidst interruptions (36.1% on Scale AI's Audio MultiChallenge). It features improved tonal understanding, dynamic response adjustment to user emotions, and inherent multilingual capabilities for real-time, multimodal conversations. All audio output is watermarked with SynthID.
  5. Improved Gemini audio models for powerful voice interactionsThis post details updates to Gemini 2.5 Flash Native Audio, enhancing its performance for live voice agents with improved function calling (71.5% on ComplexFuncBench Audio), instruction following (90% adherence), and conversational flow. It also introduces live speech-to-speech translation capabilities supporting over 70 languages, preserving speech nuances and handling multilingual input and noise.