This project aimed to develop speech generation technology that can be used to automatic creating of child stories. The generation of audible stories is based a series of neural network modules, which are trained with many hours of recordings of Cantonese child stories. Users can use a mobile app (iOS or Android) to listen to pre-loaded stories, suggest modifications on the content of stories and listen to the modified stories instantly.
Neural text-to-speech (TTS) has been investigated extensively with good success. However, existing systems are not able to generate speech with desired expression style, e.g., storytelling, conversation. The main technological challenge of this project is the modeling of multi-facet style changes in child storytelling and an effective mechanism of controling speaking style.
Founded in 1963, The Chinese University of Hong Kong (CUHK) is a forward-looking comprehensive research university with a global vision and a mission to combine tradition with modernity, and to bring together China and the West. CUHK teachers and students hail from all around the world. Four Nobel laureates are associated with the university, and it is the only tertiary institution in Hong Kong with recipients of the Nobel Prize, Turing Award, Fields Medal and Veblen Prize sitting as faculty in residence. CUHK graduates are connected worldwide through an extensive alumni network. CUHK undertakes a wide range of research programmes in many subject areas, and strives to provide scope for all academic staff to undertake consultancy and collaborative projects with industry.