The conference was chaired by Dan Ellis from Columbia University and Youngmoo Kim from Drexel University where the event took place. Apart from Sunday which kicked off with some very interesting 3-hour tutorials, the daily routine consisted of an exciting mix of panel discussions, plenary sessions and poster presentations. The relatively small number of attendees made it easy to get in touch with other researchers and get to know each other. I want to thank all volunteers, reviewers, committee members and other people who have helped to make this event so well-organized and substantial.
Over 100 papers had been chosen by the program committee to be presented at the conference. They covered an incredibly wide range of topics; here is an overview:
- User interfaces, visualization, music organization
- Music analysis (harmony, melody, timbre, rhythm, etc.) and transcription
- Programming languages for music analysis and synthesis
- Music recommendation and recognition
- Social networks, metadata, tags and blogs
Apart from many interesting poster discussions, the most exciting presentations for me where “Hit Song Science is Not Yet a Science” by Francois Pachet and Pierre Roy and the “Support for MIR Prototyping and Real-Time Applications in the ChucK Programming Language" by Rebecca Fiebrink, Ge Wang and Perry Cook.
In the first mentioned, the researches took an extensive set of musical pieces, each annotated with a set of tags categorized as acoustic, subjective or facts and used their audio classification technology to make the machine learn these features. What was really exciting was that they concluded that semantics and grounding were often not related – many acoustic features proved heard to learn while other, more subjective ones yielded better results. However, their final goal – to check if the characteristics of a popular hit song could be determined by a machine purely based on the acoustic features – proved to be unsuccessful. I think, this is not the worst of news – especially for musicians. Socio-cultural aspects also play a huge part – not to mention the record companies’ ad budget...
The other really exciting presentation was ChucK: This real-time audio programming language starts to get more and more interesting and the two presenters managed to give the audience an introduction to the language, write live code that generated polyphonic sound and demonstrate real-time training and classification of two types of classical music – all of this in less than 20 minutes. I was very impressed! This piece of software manages to make something that is usually very dry, complicated and scientific almost playful. Soon the plugin interface will be out – I can’t wait to look into this further. Kudos to these guys at Princeton University!
So, you see, this has been a very interesting week. We here at mufin are also looking forward to contribute to this scientific community in the coming years. There is a lot of cutting-edge stuff in our lab waiting to be released!
