CLASI
High quality, human like simultaneous interpretation system
- Adopting a data-driven read-write strategy to balance translation quality and latency.
- Multi modal retrieval module enhances the translation of specific domain terms.
- Generate fault-tolerant translations using Large Language Models (LLMs).
- Consider translating input audio, historical context, and retrieval information.
- In real-world scenarios, the effective information ratio (VIP) is significantly higher than in other systems.
- Support translation directions from Chinese to English and from English to Chinese.
Product Details
CLASSI is a high-quality, humanoid simultaneous interpretation system developed by the ByteDance research team. It balances translation quality and latency through novel data-driven read-write strategies, enhances translation of domain specific terms using multimodal retrieval modules, generates fault-tolerant translations using large language models (LLMs), and considers input audio, historical context, and retrieval information. In real-world scenarios, CLASI achieved effective information ratios (VIP) of 81.3% and 78.0% in both Chinese English and English Chinese translation directions, far exceeding other systems.
