The 6th Taiwanese Music and Audio Computing Workshop will take place at the Institute of Information Science on 17 April 2019. The theme this year is Music, Embodiment, and Artificial Intelligence.

We are honored to invite Masataka Goto, a world leading researcher on MIR, to this workshop. We also invite domestic speakers from AS, NTHU, and NCKU to share recent MIR studies in Taiwan.

15:30-15:55

Li Su

Automatic music transcription: state of the art and recent development

16:15-16:40

Yu-Fen Huang

The embodiment of musical mind: Decomposing musical conducting movement

Abstract

Music conductors serve to contribute their musical interpretations and coordinate musicians’ playing in orchestra. During orchestral conducting, conductors use their body movement to guide expressive variations in music. The conducting pedagogy has summarized a stock of conventionalized “conducting vocabulary” – movements regarded as carrying specific musical instructions. However, in real instances of conducting, conductors’ movement appears to be flexible, and the variations in their movement are far more beyond the conventional descriptions in pedagogy.

We therefore seek to identify expressive cuing from different conductors’ movements. The motion capture technology was applied to collect conductors’ body movement, and interviews were conducted to derive conductors’ musical interpretations. Multiple analysis techniques were adopted to extract the key body parts, key kinematic features, and to identify different types of expressive cuing in conducting. The analysis results indicate that the kinematics of conductors’ body movement is revealing regarding their musical intentions.

16:40-17:00

Amy Hung

Multi-task learning model for multi-pitch streaming and music style transfer

17:00-17:20

Hsueh-Wei Liao

Lip Buzzing Modeling with Physical-informed Neural Network

Abstract

Synthesizing the music instrument sound by physical modeling has is an approach without the requirements of training data. In contrast to physical modeling, supervised-learning based method (I.e., neural network) generate audio samples by learning the feature from the recordings instead of the physics of the sound production.

Recently, some studies have shown the potential of neural network on physical simulation. In this project, we concentrate on the simulation of the lip buzzing, which is the important excitor of the brass instrument. One of the physical models of lip buzzing was introduced by Adachi. Adachi’s model is a two-dimensional model which preforms both swinging and stretching motion. Our goal is a generative neural network for the lip buzzing sound. The architecture of this neural network is physically-informed by Adachi’s model. We hope such kind of specially designed neural network can help improve the ability of generalization. That is, the neural network can generate higher quality audio with wider range of vibrating frequency.

17:20-17:40

邱晴瑜

Expressive violin synthesis and concert of virtual violinist

17:40-18:00

蕭佑丞

Musical Phrasing Based on GTTM Using BLSTM Networks

LOC

Li Su 蘇黎 | lisu [at] iis.sinica.edu.tw

Last Update: 2019-04-23

MCTLab