What is TicVoice 7.0
TicVoice 7.0 is the seventh-generation high-quality TTS (text-to-speech) engine launched by Mobvoi, based on a new-generation speech generation model Spark-TTS . TicVoice 7.0 is based on the innovative BiCodec encoding method, decomposing speech into Global Tokens and Semantic Tokens, achieving precise control of timbre and semantics, and is highly unified with the structure of text LLMs. The engine features 3-second voice cloning capability, supports multi-role, multi-emotion, all-age groups, and Chinese-English switching, with natural and fluent sound close to broadcast-level quality. TicVoice 7.0 is already available onMoyin Gongfang The "3s voice cloning" feature is launched, widely applicable to fields such as intelligent customer service, audiobooks, and film and television dubbing, bringing users the ultimate AI dubbing experience.
Main Features of TicVoice 7.0
- 3-second voice cloning: Captures the user's voiceprint in 3 seconds, precisely replicating personalized timbre, and supports low-quality audio input.
- Multi-role and multi-emotion performance: Supports simulation of various emotions such as happiness, anger, and sadness, enhancing content expressiveness.
- All-age voice adaptation: Covers diverse timbres from children to the elderly, meeting the needs of different scenarios.
- Flexible Chinese-English switching: Supports mixed Chinese-English speech synthesis, facilitating multilingual content creation.
- Broadcast-level voice quality: Synthesized speech is clear, fluent, natural, and pleasant, with strong timbre and emotional expressiveness, approaching professional broadcast standards.
- Customized exclusive voice: Users can customize exclusive timbres according to their needs, meeting personalized dubbing requirements.
Technical Principles of TicVoice 7.0
- Innovative speech encoding method: Based on BiCodec technology, decomposing speech into Global Token (global features, such as timbre) and Semantic Tokens (semantic-related features, 50 tokens/second), balancing global controllability and semantic relevance. It solves the problems in traditional speech encoding where semantic tokens are difficult to precisely control timbre and acoustic encoding relies on multiple codebooks.
- Unified with text LLMs structure: Reuses the architecture of Qwen2.5 , based on attribute tags (such as gender and fundamental frequency level) and fine-grained attribute values (such as precise fundamental frequency), using text + attribute tags as input, sequentially predicting fine-grained attribute values → Global Tokens → Semantic Tokens. It achieves high consistency between speech token modeling and text token modeling.
- Single-stage, single-stream generation: Uses a language model (Sequence Monkey) to achieve TTS generation in a single-stage, single-stream manner, without the need for additional generative model assistance, improving generation efficiency and controllability.
- Deep learning-based speech synthesis: Based on deep learning technology, combined with large amounts of speech data to train models, achieving natural and fluent speech synthesis effects.
Project Address of TicVoice 7.0
- Project Official Website：Moyin Gongfang
Application Scenarios of TicVoice 7.0
- Intelligent customer service: Provides natural and fluent voice interaction capabilities for online customer service systems, improving user experience and reducing labor costs.
- Audiobooks and podcasts: Rapidly generates high-quality audiobook and podcast content, supports multi-role and emotional expression, enhancing listeners' immersion.
- Film and television dubbing and narration: Efficiently complete voiceover and commentary work for film, television, and short videos, support multilingual switching, and reduce production costs.
- Emotional Live Streaming and Interaction: Simulate real emotions during live streaming, enhance interaction between streamers and viewers, and increase content appeal.
- Education and Training: Provide vivid voice teaching content for online education platforms, support multiple languages and multiple roles, and improve the learning experience.