5/6(三) 論文研討─資訊科學與工程的演講，歡迎聽講！

演講者：蘇黎/中央研究院 資訊科技創新研究中心 博士後研究員

時間：5/6 (三) 15:30-17:20

地點：工程四館 117教室 (ED117)

主持人：王國禎教授

主題：Sparse Representations for Music Information Retrieval

演講摘要(Abstract)：

Using sparsity as a constraint in representing and analyzing audio signals has been found advantageous in various audio-related researches in recent years. A sparse representation of audio signals aims at representing the information of an input signal using only a small number of elementary atoms (codewords) selected from an audio dictionary (codebook). Comparing to conventional audio features, a sparse representation is characteristic of its ability of symbolizing any local audio event as a codeword, its flexibility of using an arbitrary large number of codewords learned from a corpus of audio data in an unsupervised fashion, and its robustness to outliers, noise or corruptions in the input data. Over the past few years, sparse representations of audio signals have led to competitive performance in diverse tasks such as noise reduction, compression, speech enhancement, audio watermarking, sound and music classification, melody transcription and blind source separation, to name a few. This talk aims to provide an introduction of music information retrieval (MIR), to present a few state-of-the-art research results based on sparse representations in this emerging field, and to address the growing interests in sparse representation for audio signal processing.

Biography：Li Su was born in Kaohsiung city, Taiwan. He received double B. S. degree in Electrical Engineering and Mathematics from National Taiwan University in 2008, and Ph.D. degree in Communication Engineering from National Taiwan University in 2012. Since May 2012, He has been a Postdoctoral Research Fellow in Research Center for Information Technology Innovation, Academia Sinica, Taiwan. His research projects now include musical instrument recognition, singer identification, onset detection and multiple pitch estimation (MPE). Recently my published works are mostly about using phase information and sparse coding (SC) techniques for MIR problems.