讲演题目

Automatic Speech Recognition: Challenges in Noise Robustness and LVCSR

讲演者

Qifeng Zhu（朱奇峰）

International Computer Science Institute, affiliated with UC Berkeley

讲演时间

12/16 afternoon，2004

讲演地点

东主楼10-307

？

讲演摘要

Dr. Qifeng Zhu worked on noise robustness in Automatic Speech Recognition (ASR) in his Ph.D. research at UCLA and joined the Aurora 2 evaluation in Eurospeech 2001. Now he is working on MLP-based novel features for large vocabulary continuous speech recognition (LVCSR) through the teamwork of Novel Approaches project at ICSI sponsored by DARPA. Both research has been evaluated high. The results presented in Eurospeech01 on noise robustness was evaluated as among the top 5, and the MLP based feature for LVCSR has been regarded as the top 2 technical improvement in the NIST evaluation in 2004. He is also involved in the European research project on meeting recognition and structural analysis as the AMI project. This talk will cover his research efforts in these areas and also introduce related recent technical improvements and research activities in these fields.

讲演者简介

Dr. Qifeng Zhu (B.S. 1994, M.S. 1997, Ph.D. 2001) is an active researcher in the field of automatic speech recognition (ASR). He has been working on noise robustness and large vocabulary continuous speech recognition in several frontier research projects in this field, and had 25 publications. He is a frequent invited reviewer for several IEEE sponsored journals and conferences in the field of signal processing and machine learning. His current research interests include small foot-print embedded ASR and distributed ASR, and general theories on signal processing/feature extraction and machine learning