This talk will introduce a simple tree-based method for randomly generating new audio signals in FAUST. FAUST is a functional programming language for the development of audio synthesis and DSP programs, as it can be compiled using a wide variety of audio platforms and architectures. We will then provide a live demonstration of the results in Pure Data (a visual audio programming language), then briefly explain its potential applications in future research.

Common audio spatializers use a limited amount of Head-Related Transfer Functions (HRTFs), they are mainly based on the capture of these filters by using a single listener model hindering their personalization. However, HRTFs are unique for each person since they depend on the pinnae which are also unique. A mismatch between a listener and HRTF pinnae often yields poor performance in the estimation of the location of a sound source. Capturing HRTFs for each user is not always possible since it requires anechoic chambers, specialized equipment, and prolonged sessions. To ameliorate this problem, we present the personalization of HRTFs based on the parametrization of latency space variables extracted from many HRTF databases by using autoencoders (a kind of Artificial Neural Network). With this research, we expect to increase the subjective location accuracy of sound sources in VR/AR environments.

The goal of this research is to create a real-time spatialization system based on inverse filtering and loudspeaker grouping. Currently, stereophonic sound is used in a variety of products such as video games, movies, and music technology by means of headphones. However, headphones have some disadvantages, such as not being able to share the sound with others, and being worn, it is easy for listeners to get tired. Therefore, this research seeks an implementation of stereophonic sound using loudspeakers. Past research has led to the development of various systems for stereophonic sound by loudspeakers. However, each of these systems has its own advantages and disadvantages. A new and original sound system was also developed in our lab. This sound system is done with a small number of loudspeakers, and the sound image does not collapse as the listener moves around. However, it is computationally intensive and can currently only be run offline with MATLAB. To do this in real-time, this research use an application called Pure-data.

Strangely enough, the way we hear sounds differs from person to person. For example, when we hear a single sound, whether it is “loud” or “quiet” varies from person to person. Furthermore, one experiment has shown that women are more satisfied with low volume than men. The reason why we hear differently from person to person is also related to the functioning of the brain. Psychoacoustic and psychology studies the relationship between the brain and sound by measuring and collecting such data.

In the field of music generation using machine learning, the creation of new music is currently a major challenge due to the underdevelopment of data sets and a large amount of previous research in accordance with music theory. I would like to use machine learning to create novelty in music as my research theme in graduate school. This presentation will introduce the fundamentals of music generation using opportunity learning and define the direction of my research theme by picking up issues and underdeveloped areas in the field and comparing them with my interests.

Tidalcycles has been developed for live coding of music. Tidalcycles uses Supercollider, an environment and programming language, to make sounds. In addition, we can visualize music that is created on Tidalcycles since Supercollider has the function, OSC(Open Sound Control). In this research, we will create music and do its visualization.

This talk is about the development of a wordbook application using 13 vowels and 24 consonants to improve English pronunciation by Japanese speakers. The International Phonetic Alphabet (IPA) uses special symbols and naive Japanese speakers usually find them difficult to understand, hindering the improvement of their pronounciation. As an alternative, we propose using phonetic symbols put forward by Prof. Nogita of CLR (University of Aizu)

This talk first presents some challenges and prospects awaiting M.Sc. and Ph.D. candidates and segues into the presentation of a series of techniques and hacks that could ease your life as a researcher. I will briefly introduce inflationary tendencies in education, how to keep your own bibliography up to date.