What began as a seemingly innocuous WeChat initiative to record people in their native dialects so as to improve voice recognition capabilities, has acquired sinister overtones.
WeChat had invited people to read everyday phrases in their local dialects, which are recorded and the payment is made on the basis of one yuan per three sentences recorded. Some people, according to various accounts, made a few hundred yuan.
WeChat's aim was to better understand regional accents and linguistic variations, thereby improving the accuracy and inclusivity of its speech-to-text systems. But could the same data that enhances user experience also strengthen the platform's ability to monitor voice communications?
In a tightly regulated digital environment like China, technological advancements in voice analysis could also expand the scope of surveillance.
Research by Citizen Lab, which studies internet openness and security from its base in Toronto, noted that content shared on WeChat including messages, images, and files can be scanned and analysed, often without users being aware. Voice identification and content analysis also become easier using current technology.
According to the Australian Strategic Policy Institute, advances in artificial intelligence are transforming China's state control mechanisms into increasingly precise tools for managing populations and targeting groups domestically and abroad.
While AI-powered visual surveillance is already well documented, the report highlights how large language models and related systems are being used to automate censorship, enhance monitoring, and pre-empt dissent.
More recently, China has also developed language-specific AI systems, including models designed for the Tibetan language, such as DeepZang.
WeChat is already known to block the use of certain words and phrases close to anniversaries such as the Tiananmen Square protests of 1989, for instance.