报告题目：Deep Learning for Audio Classification

报 告 人：Prof. Wenwu Wang

报告时间：12月10日（周二 10:00~11:30）

报告地点：航海学院323会议室

邀 请 人：于洋 副教授

报告摘要：

Audio classification (e.g. audio scene analysis, audio event detection and audio tagging) have a variety of potential applications in security s             urveillance, intelligent sensing for smart homes and cities, multimedia search and retrieval, and healthcare. This research area is under rapid development recently, having attracted increasing interest from both academia and industrialists. In this talk, we will present some recent and new development for       several challenges related to this topic, including data challenges (e.g. DCASE challenges), acoustic modelling, feature learning, dealing with weakly        labelled data, and learning with noisy labels. We will show some latest results of our proposed algorithms, such as the attention neural network algorithms for learning with weakly labelled data, and their results on AudioSet – a large scale dataset provided by Google, as compared with several baseline methods. We will also use some sound      demos to illustrate the potentials of our proposed algorithms.

报告人简介：

Wenwu Wang is a Professor in Signal Processing and Machine Learning, and a Co-Director of the Machine Audition Lab within the Centre for Vision Speech and Signal Processing,     University of Surrey, UK. He has been a Senior Area Editor (2019-) and Associate Editor (2014-2018) for IEEE Transactions on Signal Processing. He was a Publication Co-Chair for           ICASSP 2019, Brighton, UK, and will serve as Tutorial Chair for ICASSP 2024, Seoul, South Korea. His current research interests include blind signal processing, sparse signal processing, audio-visual signal processing, machine learning and perception, artificial intelligence, machine audition (listening), and statistical anomaly detection. He has (co)-authored over 250             publications in these areas.

More information on his personal page:

http://personal.ee.surrey.ac.uk/Personal/W.Wang/