Shujie Liu
Principal Researcher at Microsoft Research Asia
Shujie Liu is a Principal Researcher at Microsoft Research Asia, graduated from Harbin Institute of Technology in 2012, and joined Microsoft Research Asia in 2012, working on natural language processing, speech processing, and machine learning. He has published more than 100 papers in top journals and conferences on natural language processing and speech processing, co-authored the book "Machine Translation", and participated in writing the book "Introduction to Artificial Intelligence". He has won several first places in international natural language and speech processing evaluation competitions. He has served as a reviewer and field chair for several international conferences. His research results are widely used in important Microsoft products such as Microsoft Translator, Skype Translator, Microsoft IME and Microsoft Speech Services.
Topic
Exploration and application of pre-training methods in the field of speech processing
With the progress of pre-training models in natural language processing and image processing, speech pre-training models have also received increasing attention. Speech processing methods based on pre-training models have also made breakthroughs in several tasks. In this report, we will introduce three different pre-training models, encoder-based pre-training model, pre-training model based on encoder-decoder framework, and pre-training model based entirely on decoder, respectively. The first two pre-training models are mainly applied to tasks related to speech recognition and speech understanding, while the fully decoder-based language model is mainly applied to high-quality speech generation tasks. Based on this, we further explore the future research direction of pre-training methods in the field of speech processing.