A community is driven by its long-term goals, but at present such goals for music information retrieval are not well-defined. The vision community has the DARPA Urban Challenge, the speech community has speech-to-speech transcription, and the artificial intelligence community has reproducing the brain. Some great challenges of the past few years that have triggered new research are recommendation and automatic tagging. Once these are solved what else will we turn to?
For example, is it possible to perform audio remixing without intermediate stages such as transcription or chord recognition? Can a model understand musical forms well enough to compose? Could such a model pass a "musical Turing test"?
Why is stating such goals important? First, to draw the most promising students to our field, we must be able to challenge them with “transformative” ideas. Second, such ambitious and (hopefully) interdisciplinary activity helps us to reach outwards to other research communities.
No student comes into this field in order to implement the best beat tracking system or an improved midi alignment algorithm. We do not imply that these tasks are not of value: they have aided in our community development by increasing our knowledge and preparing us for more difficult tasks. However, by themselves, they are not end goals.
Music Information Retrieval is in need of a place to shape these goals as a community, and the f(MIR) workshop is intended to address this need.
Thierry Bertin-Mahieux, Amélie Anglade, Jason Hockman - f(MIR) Organizers