The latest voice recognition system, ASR, launched by Tencent Cloud has been upgraded to enhance its capabilities in handling dialects and noise. ASR supports the recognition of 23 regional dialects and can manage mixed dialect situations. Through its proprietary hybrid recognition engine, ASR can quickly identify and switch between different dialects. Employing distillation algorithms and semi-supervised training, ASR efficiently learns from both labeled and unlabeled data, even handling small minority dialects. In terms of semantic recognition, ASR utilizes large language models to extract intentions through contextual semantics and big data, improving recognition accuracy. Additionally, ASR has developed noise-resistant capabilities through self-learning and optimization algorithms, automatically filtering out noise to enhance recognition precision. ASR has been widely applied within Tencent's internal products and various industries outside, serving thousands of enterprise clients with daily call volumes reaching billions.