Llasa-3B
| Property | Value |
|---|---|
| Author | HKUSTAudio |
| Model Size | 3 Billion Parameters |
| License | CC BY-NC 4.0 |
| Training Data | 250,000 hours Chinese-English Speech |
What is Llasa-3B?
Llasa-3B is an innovative text-to-speech (TTS) system that builds upon the LLaMA language model architecture. It integrates XCodec2 codebook containing 65,536 speech tokens, enabling high-quality speech synthesis in both Chinese and English. The model represents a significant advancement in neural TTS technology, seamlessly incorporating speech generation capabilities into the LLaMA framework.
Implementation Details
The model utilizes a sophisticated architecture that combines LLaMA's language understanding capabilities with speech token generation. It converts audio into single-codebook tokens, treating speech synthesis as a language modeling task. This approach enables compatibility with existing LLM optimization techniques, including compression, acceleration, and fine-tuning methods.
- Integrated XCodec2 codebook with 65,536 tokens
- Supports both direct text-to-speech and speech-prompted synthesis
- Compatible with LLaMA framework optimizations
- 16kHz speech output support
Core Capabilities
- Direct text-to-speech synthesis
- Speech-prompted generation maintaining voice characteristics
- Bilingual support (Chinese and English)
- Configurable generation parameters (temperature, top-p sampling)
Frequently Asked Questions
Q: What makes this model unique?
Llasa-3B's unique approach lies in treating speech synthesis as a language modeling task, enabling seamless integration with LLM frameworks while maintaining high-quality speech output. The model's ability to handle both direct text input and speech prompts sets it apart from traditional TTS systems.
Q: What are the recommended use cases?
The model is ideal for applications requiring high-quality bilingual speech synthesis, including voice assistants, content creation, and accessibility tools. However, due to its CC BY-NC 4.0 license, it's restricted to non-commercial applications.
