Breeze TTS 2 - BreezeBlue's next-generation speech synthesis model

What is Breeze TTS 2?

Breeze TTS 2 is a next-generation voice synthesis model from BreezeBlue. It supports zero-sample design of character voices using natural language and precise control of performance details such as emotion, speech rate, and accent using natural language commands. The model-side latency is as low as 40ms, ranking first globally in three authoritative evaluations: Voice Design, Voice Direction, and Latency. It supports 50+ languages.

Main features of Breeze TTS 2

• Tone DesignIt creates unique character voices from scratch by describing age, temperament, personality and other characteristics using natural language, without relying on a preset voice library.
• Voice guidance: Use natural language commands to precisely control emotions, speech rate, accent, physiological state and dynamic performance changes, while maintaining consistency between the character's voice and identity.
• Ultra-low latency real-time generationThe model-side first-byte latency is as low as 40ms, and it supports WebSocket streaming to meet the needs of real-time dialogue and interaction.
• Multilingual coverageSupports over 50 languages, maintaining naturalness of voice, consistency of character, and performance control in cross-language scenarios.

Technical Principles of Breeze TTS 2

• Paradigm Transfer of Large Language ModelsBreeze TTS 2 follows the technical approach of LLM, transferring the generalization ability, instruction compliance ability and Scaling Law of language models to the field of speech generation, enabling speech models to understand complex character settings and contexts.
• Long conversation-oriented data strategyThe team collects real long dialogue data from movies, TV series, interviews, and podcasts, focusing on character relationships, emotional tension, conversational rhythm, and pause intentions. This allows the model to learn how humans express themselves in complex interactions, rather than through standardized text readings.
• Reconstruction of Scenario-Driven Evaluation SystemAfter discovering that academic benchmarks were out of touch with the actual needs of creators, the team rebuilt the evaluation criteria around the two core aspects of timbre design and sound directing, and open-sourced the corresponding benchmarks, using cross-scene role consistency and continuous change capability as the basis for model iteration.
• End-to-end unified architectureIt integrates timbre design, sound guidance, and low-latency streaming generation into a single model, and achieves character creation, performance control, and real-time response through a unified inference chain, avoiding the latency and consistency loss caused by splicing multiple modules.

How to use Breeze TTS 2

• BreezeCreator Web VersionVisit the BreezeCreator platform website https://breezeblue.ai/app, use natural language to describe the character's characteristics to design a unique voice, and then use natural language commands to adjust emotions and performance to directly generate audio content.
• HTTP API callsAfter obtaining the API Key, developers can submit text and tone/performance parameters through a standard HTTP interface to obtain synthesized speech files, which is suitable for batch content production.
• WebSocket Real-Time APIEstablish persistent WebSocket connections, append text in real time, and stream raw PCM audio, suitable for low-latency interactive scenarios such as virtual anchors, game NPCs, and AI agents.
• SDK IntegrationIt can be quickly integrated using the official JavaScript/TypeScript SDK, allowing you to create real-time voice conversations, manage dialogue turns, and play audio with just a few lines of code.

The core advantages of Breeze TTS 2

• Zero-sample tone designIt creates unique character voices from scratch by describing age, temperament, personality and other characteristics using natural language, without relying on a preset voice library.
• Natural Language Voice Guidance: Use natural language commands to precisely control emotions, speech rate, accent, physiological state and dynamic performance changes, while maintaining consistency between the character's voice and identity.
• Ultra-low latency real-time generationThe model-side first-byte latency is as low as 40ms, ranking first in the world in TTS Latency Benchmark, meeting the requirements of real-time interaction.
• Multilingual role consistencySupports 50+ languages, maintaining naturalness of voice, consistency of character, and performance control in cross-language scenarios.
• Leading in all aspects of the evaluationIt ranks first in the world in three authoritative evaluations: Voice Design, Voice Direction, and Latency.

Breeze TTS 2 project address

• Project official website：https://breezeblue.ai/breeze-tts-2

Comparison of Breeze TTS 2 with similar competing products

Application scenarios of Breeze TTS 2

• Virtual anchors and AI live streamingAI anchors adjust their tone and perform in real time based on the comments, and a single live stream can last for 1-2 hours, with hundreds of live streams completed to date.
• Game NPCs and Role-PlayingGenerate unique voices for thousands of NPCs in the game, supporting real-time dialogue and emotional changes to enhance immersion.
• Audio content and radio dramasCreators use natural language to design voices for multiple characters and produce long-form radio dramas (such as fan-made versions of "The Three-Body Problem"), which become popular on the first day of release.
• AI Agent and Intelligent Customer ServiceLow-latency real-time voice interaction enables AI Agents to proactively report progress, accept interruptions, understand user intent, and continue to collaborate.
• Interactive entertainment and virtual companionshipIt gives virtual characters and AI companions unique voice identities, supporting long-term, role-based emotional interaction and story experience.