IntermediateAI

Google Releases Gemini 3.1 Flash TTS with 200+ Audio Tags

On April 15, 2026, Google DeepMind released Gemini 3.1 Flash TTS — a text-to-speech model that introduces more than 200 granular audio tags for steering vocal style, tone, pacing, and accent, and tops the Artificial Analysis TTS leaderboard with an Elo score of 1,211. The preview is available through the Gemini API, Google AI Studio, Vertex AI, and Google Vids, with support for 70+ languages and native multi-speaker dialogue.

Gemini 3.1 Flash TTS announcement hero image
Image credit: Google

What’s New

Gemini 3.1 Flash TTS is Google’s most expressive speech model to date, focused on controllability rather than raw scale. Developers can embed natural-language audio tags directly into the input text — covering delivery style (whispering, excited, calm), pacing (fast, slow), accent, and even scene direction for multi-speaker scripts. Tags can be placed inline mid-sentence to shift expression on the fly, and speaker-level tags let a single prompt produce a dialogue with distinct voices without separate API calls.

The model supports more than 70 languages and is positioned by Google as offering a strong quality-to-cost ratio. The paid tier is priced at $1.00 per million input tokens and $20.00 per million audio output tokens, with a batch mode offering a 50% discount. A free tier is available for experimentation, though Google notes that free-tier data may be used for product improvement.

Benchmarks

On the Artificial Analysis TTS leaderboard — a blind human-preference evaluation — Gemini 3.1 Flash TTS scored 1,211 Elo, placing it among the top entries for expressive speech synthesis. Google highlights its position in what the leaderboard calls the “most attractive quadrant” for combined quality and cost.

Gemini 3.1 Flash TTS benchmark evaluation chart
Image credit: Google

Safety and Availability

Every audio clip generated by Gemini 3.1 Flash TTS is watermarked with SynthID, Google’s imperceptible identifier for AI-generated content. The watermark is designed to survive common audio transformations while preserving audible quality, giving platforms a way to detect and label synthetic speech.

Developers can access the preview via the Gemini API and Google AI Studio; enterprises get it through Vertex AI; and Google Workspace users can use it inside Google Vids for narration and voiceovers. The release was led by Vilobh Meshram (Senior Product Manager) and Max Gubin (Principal Research Engineer) on the Google DeepMind speech team.

What This Means

Controllable TTS is becoming the competitive frontier. With Mistral’s Voxtral TTS and Alibaba’s Qwen3-TTS targeting open-weight deployments, and ElevenLabs defending the commercial voice market, Google’s play is to bundle expressive control directly into the Gemini API surface developers already use for text and multimodal work. The 200+ audio tags make Gemini 3.1 Flash TTS particularly well-suited to audiobook production, video dubbing, and conversational agents where emotional range matters more than sheer naturalness.

This post was drafted with AI assistance and reviewed by RITS staff.

Sources

Related stories

agentic ai

Google Releases Gemini 3.1 Pro with 2× Reasoning Performance

Key points Google DeepMind released Gemini 3.1 Pro on February 19, 2026, describing it as its most capable model to date and the first in the Gemini family to receive a “.1” mid-cycle update rather than the usual “.5” increment. Gemini 3.1 Pro scores 77.1% on ARC-AGI-2, which the article says is more than double […]

AI

Google Launches Gemini Embedding 2: Its First Multimodal Embedding Model

On March 10, 2026, Google DeepMind released Gemini Embedding 2 — the company’s first natively multimodal embedding model. Unlike traditional embedding models that handle only text, Gemini Embedding 2 maps text, images, video, audio, and documents into a single unified vector space, enabling cross-modal search, classification, and clustering across over 100 languages. Intermediate Key points […]

AI

Voxtral TTS: Mistral’s Open-Weight Text-to-Speech Model Rivals ElevenLabs

On March 26, 2026, Mistral AI released Voxtral TTS — a 4-billion-parameter open-weight text-to-speech model that the company says outperforms ElevenLabs Flash v2.5 in human preference tests while matching ElevenLabs v3 in lifelike interactions. Built on Ministral 3B, Voxtral TTS supports nine languages, clones voices from just three seconds of audio, and achieves 70ms latency […]

Never miss a RITS update

Subscribe to the RSS feed for AI, emerging technology, and digital scholarship news curated for the NYU Shanghai community, or get in touch to suggest a story.