(Yicai) May 28 -- China Telecom, one of the nation's largest wireless carriers, unveiled China's first large language model that can simultaneously understand over 30 local dialects, including Cantonese, Shanghainese, and Wenzhounese, to make customer service more efficient.

The speech recognition model, which was launched last weekend, has been tested in customer service in Fujian, Jiangxi, and Guangxi provinces, according to a press release by an artificial intelligence lab of the Beijing-based company. The model can understand 30 dialects in a matter of seconds and handle about two million calls per day.

Moreover, some local governments have been using the model in public services to enhance customer service.

The LLM is open-source as the goal is to protect local dialects with partners and individual developers, China Telecom said. The company will continue to improve the number and accuracy of supported dialects. The plan is to build the first speech recognition LLM that supports dialects of 333 prefecture-level regions, as well as major minority languages across China.

The model is of great social significance because it is extremely urgent to protect dialects to pass on language and culture, the firm said. China has five major language families and over 130 languages. About a fifth of the population does not understand Mandarin, a group of residents who often live in rural areas and are isolated from computers. Moreover, many dialects are dying out due to China's rapid socio-economic development, it added.

Editor: Emmi Laine