Weibin Zhang has recently joined our speech researchers. Weibin is a PhD candidate at the Hong Kong University of Science and Technology, where he has studied on a Hong Kong PhD Fellowship since 2010. Before that, he worked for Datang Mobile, first as a researcher and later as team leader of the multimedia group. While at Datang, he worked on speech enhancement algorithms such as acoustic echo cancellation and also on a multimedia system, Eden. The system uses OpenMax IL, an application programming interface, to enable application-layer integration of a speech coder, a video coder, and speech enhancement methods uniformly.

His thesis work focuses on low-resource speech recognition. Some languages - for example, Cantonese - have limited training data to train speech recognition systems. His work focused on acoustic models with Gaussian mixture models as the output distribution of feature vectors. More specifically, to explore ways to deal with the over-fitting problem associated with low-resource acoustic modeling, Weibin has been exploring regularization methods, especially lasso regularization that leads to acoustic models with sparse inverse covariance matrices. At ICSI, he will be exploring the use of deep neural networks in speech recognition while working on the IARPA-funded Swordfish project. He is more generally interested in machine learning, and he has also worked on image processing. In his spare time, Weibin enjoys playing badminton and swimming.