On May 26th, China Telecom's Artificial Intelligence Research Institute unveiled the industry's first large-scale speech recognition model that supports free mixing of 30 dialects - the Starry Multidialect Mixed Speech Large Model.
This large model addresses the pain point of single models being able to recognize only specific individual dialects. It can simultaneously recognize and understand over 30 dialects including Cantonese, Shanghainese, Sichuanese, Wenzhounese, making it the largest dialect-supporting speech recognition model domestically.
According to reports, the research and development team has built a high-quality dialect database with over 30 dialects and more than 300,000 hours.
This system not only significantly reduces the bit rate of speech transmission during inference but also makes communication more natural and smooth, addressing the issue of information services being inaccessible to the elderly and "poor and remote" areas.
It is worth mentioning that He Zhongjiang, the general manager of China Telecom Artificial Intelligence Technology Co., Ltd., stated that the algorithm and training code for the large-scale speech model will be open-sourced to the public.
Reportedly, the Xingchen speech model has been piloted in China Telecom's 10,000 intelligent customer service centers in Fujian, Jiangxi, Guangxi, Beijing, Inner Mongolia, and other regions.
After integrating the Xingchen large model, intelligent customer service can instantly understand 30 dialects, handling approximately 2 million calls per day on average.
Furthermore, the Xingchen speech model has also been implemented in the 12345 platforms in multiple cities.
