Joined Alibaba

I joined Alibaba Group in November of 2017 focusing on applying deep learning and machine learning techniques to recommender algorithms and recommender systems.  

ICASSP2017 Paper on Singing Transcription (Yamaha Internship Work)

During Oct-Dec 2015, I was a research intern at Yamaha Corporation Japan supervised by Dr. Akira Maezawa for a sung melody transcription project. We employed the novel observation probability distribution of combination of angular and gamma distributions to the HMM-based framework. The work is accepted as a proceeding of the 42nd IEEE International Conference on…

Scientific American (Chinese Edition) Interviews 环球科学采访

The Chinese edition of Scientific American,《环球科学》(“科学美国人”中文版) interviews Dr. Luwei Yang (杨璐威博士) on “Investigating the Mysteries of Music Using Quantum Physics Methods” (《用量子力学中的方法探索音乐的奥秘》)! The interview text is reproduced below.  Source:   专访 | 用量子力学中的方法探索音乐的奥秘? Original 2017-04-11 紫苏 科研圈 杨璐威博士将原本用于量子物理的谐波(harmonics)分析的方法用于自动检测音乐中的颤音。 撰文 紫苏 可能平常人看来,音乐不过是将一堆乐谱中音符转换为音频信号。但是实际上,音乐有它自己的结构和形式。对于这些音乐的特征,人们不仅可以通过定性分析(如音乐评论等),也可以利用数学方法来进行定量分析。 音乐数学建模是通向计算音乐和大规模自动化分析音乐的第一步:它不仅能让我们理解乐曲中的不同模式,也让我们理解音乐中的总体趋势。 来自的伦敦玛丽女王大学数字音乐中心的杨璐威博士将原本用于量子物理里的谐波(harmonics)分析的方法——过滤对角方法(FDM),来自动检测音乐中的颤音。单论检测准确度来说,在帧水平上,提高了30%以上;在音符水平上,提高了一倍以上。论文发表于(论文信息见文末)。 杨璐威在演奏二胡 《环球科学》的记者就这篇论文采访了杨璐威博士: 您的这篇论文的主旨是克服音乐的颤音问题是么?那么什么是颤音呢? 璐威:颤音是指音高在一定频率范围内的周期变化。它是音乐演奏中常用的技法之一,常见于弦乐、声乐、管乐中等(弦乐中的颤音也叫揉弦)。颤音一般不在乐谱中标记,因此,音乐家更能自由地让颤音出现在乐曲中的任何位置和控制颤音的快慢和大小。它能反映不同音乐文化、音乐时期、甚至是不同演奏家的音乐风格。 我们先听一段包含较多颤音的音乐片段 二泉映月, 二胡, 片段-黄江琴 (00:52) 颤音的频率一般是在4-8Hz之间。前人一般使用的方法是对基频时序信号运用短时傅里叶变换(STFT),在频谱中4-8Hz频段里找寻是否有极值的方法来判断颤音的存在。考虑到傅立叶变换的不确定性原理(uncertainty principle),如果要想在如此小的频段上寻找极值,最完美的方法就是得到一个频域分辨率很高的频谱图。并且一般情况下,需要增加窗的长度。如果窗的长度增加了,时域上的分辨率就会下降。 为什么您在研究中选择了二胡和小提琴,在二胡的颤音更为多见么?… Quantum physics offers insight into music expressivity publishes press release on our paper, “The filter diagonalisation method for music signal analysis: frame-wise vibrato detection and estimation”, just out in the Journal of Mathematics and Music. Yang, Luwei, Khalid Z. Rajab & Elaine Chew (2017). The filter diagonalisation method for music signal analysis: frame-wise vibrato detection and estimation. Journal of Mathematics and Music, pp.1-19.  Doi: 10.1080/17459737.2016.1263897…

Joined Huawei

In the February of 2017, I joined Huawei Technologies Co. Ltd. at Shenzhen China . I have done research on machine learning. Currently, I was also involved in a chatbot project with interests in intent recognition, sentiment analysis and speech emotion recognition.

Luwei Yang will attend and present the AVA at ISMIR2016

Luwei Yang will attend and present a paper about the AVA at the 17th International Society for Music Information Retrieval Conference (ISMIR2016), organised by the New York university and Columbia University. The conference will take place at the New York City 7-11 August 2016. AVA: An Interactive System for Visual and Quantitative Analyses of Vibrato and Portamento…

AVA Matlab toolbox is available

AVA is an automatic vibrato and portamento detection and analysis tool. It accepts raw audio and automatically tracks the vibrato and portamento to display their expressive parameters for inspection and further statistical analysis. Download and more information please go to here.

2016 Chinese New Year Concert

The 2016 Chinese New Year Concert will take place 7:30pm on Monday, 1 February 2016, in the Octagon and features award winning Chinese instrumentalists, and C4DM members (Katerina Kosta, Beici Liang, and Elaine Chew) and music technologies embodied in Andrew McPherson’s Magnetic Resonator Piano, and Mathieu Barthet and Gyorgy Fazekas’ Mood Conductor/Rater. Tickets will open on…

Presentation in FMA2015

From 10th to 12th June 2015, I presented a poster titled in Vibrato Characteristics and Frequency Histogram Envelopes in Beijing Opera Singing in the 5th International Workshop on Folk Music Analysis. The extended abstract is here.