Special Sessions

In addition to traditional CMMR topics, we propose a certain number of special sessions. The aim of these sessions is to encourage more specific contributions.

Submissions to these sessions are made by selecting the title of the special session in the topics section of the standard paper submission page.

Grammar and semantics for music

organized by Satoshi Tojo (Japan Advanced Institute of Science and Technology)

It is corroborated in various ways that the origin of music and language is one and the same. Therefore, we can expect an existence of constructive grammar also in music. In this special session, we focus on the study of finding grammar of music, either by statistical methods or by formal generative rules. Furthermore, we discuss what is the meaning of music, or whether music could convey mutually understandable information, as semantic aspect of music.

Music generation with AI

organized by Satoru Fukayama (National Institute of Advanced Industrial Science and Technology, AIST)

This session calls for papers related to music generation utilizing AI or machine learning. The session not only welcomes papers applying state-of-the-art AI, but also papers using previous methods for a different objective or replicating music generation experiments in previous research to obtain valuable insights. At the session, performing or demonstrating the generated music is highly recommended so that the participants could obtain an intuitive understanding of the research.   

Improvisation: its interpretation and implementation

organized by Keiji Hirata (Future University Hakodate)

Singing voice computing 

organized by Tomoyasu Nakano (National Institute of Advanced Industrial Science and Technology (AIST)) and Masanori Morise (Meiji University)

Computer-supported Interactive Systems for Music Learning

organized by Yoshinari Takegawa (Future University Hakodate) and Masaki Matsubara (University of Tsukuba)

This special session calls papers that present new computer-supported interactive systems for  learning of music (including e.g. performance, listening, composition, recording, mixing, analysis, arrangement). Contributors from a wide range of research fields focused on human-machine collaboration are welcome: for example, (but not limited to) machine learning, human computer interaction, music informatics, semantic and affective computing, mobile computing, sensor and wearables, information visualization, sound/image interaction, synthesis and sound design, and XR (VR, MR, AR). The papers are encouraged to discuss if and how the systems can transform user skill and experiences compared to traditional practices, how such skill and experiences can be evaluated and the results of the evaluations when available.

Music as/with pop-culture

organized by Ryosuke Yamanishi (Ritsumeikan University) 

Music is art, culture and entertainment content. Music has been evolved with society and technology for a long time. Especially in recent years, music is introduced into other entertainment contents: movies, anime, and games. Such pop-culture contents benefit the power of music to enrich its attractiveness. Also, the music itself is influenced by pop-culture. Song, performance and listening style has changed with technology in media processing and telecommunication. Based on the background, this session welcomes the research where music is assumed as pop-culture and music is addressed with pop-culture.

Non-standard computing for sound and music: quantum, optical, bio, etc.

organized by Eduardo Miranda

Title (tentative): Learning to Generate Music from Audio via MIR techniques

Organized by Yi-Hsuan Yang (Academia Sinica, TaiwanAI Labs)

Abstract (tentative):
Most AI models today use symbolic, score-like music data such as MIDI files to learn to compose music. While exciting progress is being made, the music generated by these models is usually not expressive enough. A main reason is that, as music performance is an artistic and subjective process, a music score mainly specifies what to be played, leaving much space for a musician to decide how to play it. Without performance-level guidance, the AI cannot generate music that is ready to be directly listened to. We explore within our group an alternative approach that learns to compose music from musical audio recordings, by capitalizing state-of-the-art music information retrieval (MIR) techniques.  Doing so gives our model opportunities to learn to generate expressive music, as performance-level cues such as variations in velocity and microtimings (timing offsets) can be found in the audio files.  I will talk about the technology that we have developed, as well as demonstrate the generated music of such AI models.