[Seminar] The Informatics Philharmonic

Christopher Raphael
Indiana University
2015년 4월 17일 금요일 AM 10:30

■ 호스트: 김선 교수 (x7280, 02-880-7280)
■ 문의: 생물정보 및 생명정보 연구실, 안홍렬 (x1784)


I will discuss and demonstrate my musical accompaniment system, the Informatics Philharmonic, which plays an orchestral accompaniment that listens to, follows, and learns from a live soloist. The system is formed of three interacting components: Listen, Predict, and Play. The Listen component uses a hidden Markov model that performs an ongoing match between the incoming audio data from the soloist and the musical score, which is known to the system. The system periodically detects that notes have occurred, treating these as observed variables in the Predict component --- a Kalman-filter-like model that governs the musical timing of the performance. Using this model, the system predicts when future accompaniment events will occur and schedules them accordingly. The Play component generates the outgoing audio by performing time stretching on a prerecorded orchestral accompaniment. This component is driven by the accompaniment times resulting from the Predict model, foll owed like a trail of breadcrumbs, which are continually changing as the model acquires new information. The Predict component is trained through a series of rehearsals to anticipate and better follow the soloist's expressive gestures and tempo changes. The talk includes a live demonstration of the system with myself playing the oboe on some short pieces or excerpts. I will also show a video that demonstrates the way the system reasons in real-time, continually setting a future course and modifying the course as the soloist plays.

연사 소개

Christopher Raphael is a professor of computer science in the School of Informatics and Computing at Indiana University, where he heads the Music Informatics program. He received his PhD in Applied Mathematics from Brown University and worked in speech and language processing at BBN before beginning his academic career. His research interests include probabilistic modelling, belief networks, and reasoning under uncertainty, with a strong focus on applications. These include recognition problems such as audio recognition, optical music recognition, as well as the modelling of musical structure and expression. Before beginning his academic career he was a professional oboist. He appeared as soloist with the San Francisco Symphony, was a fellow at Tanglewood, and soloed with the Santa Cruz Symphony on several occasions. His academic career has been driven by the goal to combine these scientific and musical interests in various ways.