CCTV News (央视新闻)

Solution

Baidu AI Cloud Xiling and CCTV News built an AI sign-language anchor that translates mainstream media content into synchronized sign-language performance for hearing-impaired audiences. The solution breaks the problem into three layers: speech must first be recognized accurately, Chinese text must then be translated into sign-language code with the right grammar and keyword structure, and finally a digital human must express the result through coordinated hand movement, mouth shape, and body language. Baidu's speech-recognition model provides the first layer, while a sign-language translation workflow was trained with extensive annotation support from deaf teachers and students at Tianjin University of Technology. On the expression layer, visual-recognition and motion-generation technology drive a 3D digital human to perform sign language with high fidelity, including precise mouth-shape generation. The same Xiling platform later expanded into a broader one-stop digital-human infrastructure, but its sign-language anchor already proved that accessibility-focused AI media can operate at broadcast scale.

Data flow

Broadcast speech is transcribed by ASR, converted from Chinese text into sign-language codes through a specialized translation engine trained with deaf-community annotations, then rendered by a digital human whose motion, mouth shape, and expression are generated for synchronized viewing by hearing-impaired audiences.