We presented CompMusic and some preliminary research results at ISMIR 2011, conference that took place in Miami, Florida (USA) from October 24th to the 28th of 2011. This is the first conference in which we have talked about CompMusic and thus it was a relevant milestone for the project. Since ISMIR is also the conference in which the research of CompMusic might fit the best I thought it was useful to report on it and give my view from our project perspective.
ISMIR is the most important conference in the field of Music Information Retrieval and it has become a quite consolidated academic conference (Downie et al., 2009). For example, this year 133 papers were presented and the acceptance rate was 59%, numbers that have been more or less stable for the past eight years, basically since we organized ISMIR in Barcelona in 2004. The review process has been improving over the years and by now the quality of the reviews is among the highest compared with the conferences I have been involved in. It is an academic conference, but at the same time it attracts the interest of quite a number of companies working in music retrieval applications, thus giving it a broader and applied context.
The research topics covered and emphasized at ISMIR have been changing over the years. Since the first conference in 2000, ISMIR has always been dominated by the audio processing papers. In the first few years there was also a significant number of papers on topics related to library sciences, musicology and symbolic-based information processing. In the last few years, only the papers on symbolic topics have maintained a relevant presence in front of the large percentage of audio processing papers. Given this situation, most of the research methodologies used come from the fields of signal processing and machine learning, thus impoverishing the potential for advancements based on more interdisciplinary research approaches. This year there were some new and interesting trends that I will comment on later.
Even though there are efforts to extend the size and diversity of the datasets used in MIR research, the music being studied by the ISMIR community comes mainly from western commercial music repertoires. For example, a highlight at this year's conference was the presentation of the freely-available Million Song Dataset, which includes contemporary popular music tracks of a limited number of repertoires. There is a clear need to put together and make available datasets with a wider variety of music repertoires.
MIREX, the Music Information Retrieval Evaluation eXchange that is coupled with ISMIR and that started in 2005, gives a good insight into the core topics being worked on and on the progress being made. Most of the MIREX tasks relate to feature extraction, such as onset detection or chord estimation, and the evaluation procedures measure the performance of the algorithms on a test dataset, again using western commercial music repertoires, compared to a ground truth that is made beforehand. In this year's campaign there were sixteen tasks being evaluated of which only one was based on symbolic data, the rest were all audio based. The focus in MIREX has always been on evaluating retrieval performance based on precision and recall approaches. This year there was a good paper by Julián Urbano identifying the challenges and opportunities of the MIREX initiative (Urbano, 2011).
We saw some possible new trends at this year's ISMIR of relevance to CompMusic. For example, there were a number of papers related to musics outside the traditionally used western commercial repertories and there were also papers on user studies, not that common in previous editions. I now try to highlight some of the papers that I found of interest for our project.
For the first time at ISMIR there was an oral session dedicated to "non-western music" (I do not like the term non-western to identify one type of music so at least we should use the plural: non-western musics). I hope this session is just the beginning of a trend that shows an increasing interest on the diversity of our world's musics. In this session we presented two papers, one that was a general introduction to CompMusic (Serra, 2011) and another one that studied the issue of tuning in Carnatic and Hindustani music (Serrà et al., 2011). The session included two other papers, one on melodic contour analysis of Jewish Torah trope performances (van Kranenburg et al., 2011) and the other on the system Tarsos (Six & Cornelis, 2011) which is a platform to explore pitch scales using pitch-frequency histograms, thus very much based on the work done by Bariş Bozkurt (Gedik & Bozkurt 2010).
There was also a poster session that included a number of papers studying different music repertories, where we presented an overview paper on the state of the art in melodic analysis of Carnatic music (Koduri et al., 2011). Sertan Şentürk, now in CompMusic, presented a paper on modeling improvisation in Turkish folk music (Şentürk & Chordia, 2011) and there was another paper on Iranian music (Abdoli, 2011).
At the level of symbolic music processing I found interesting a cross-cultural study based on the automatic clustering of different folk music traditions (Juhász, 2011). There was also a paper on the recent additions to music 21 (Cuthbert et al., 2011), an open source toolkit written in Python for analyzing, searching, and transforming symbolic music data. Despite the focus on harmonic analysis, which is not so relevant outside western tonal music, this toolkit should be quite useful for processing the symbolic representations of the musics being studied in CompMusic.
One of the research interests in CompMusic is the development of culture specific music ontologies. The paper on the issues around the development of a musical instrument ontology (Kolozali et al., 2011), work done in the context of the Music Ontology Specification, is a good reference for us, despite focusing on western instruments. This is very much related to the Linked Data initiatives and we were also very interested by the tutorial by Kevin Page and David De Roure that covered the main linked data issues of relevance to MIR.
On the last day of the conference there was a Demo Session in which we presented the new version of Freesound (Akkermans et al., 2011) with which we are starting to work on the community profiling issues of relevance to CompMusic.
The keynote by David Huron was an inspiring one, focusing on music cognition and touching on the topic of cognitive cultural diversity (Huron, 2004). He made a strong point indicating that we should worry about the loss of music culture diversity. At the industrial panel there was also some references to the need to expand our technologies to support different music cultures.
So ISMIR 2011 was a good conference. It gave me a view on the current research being done in MIR and it was interesting to see the growing interest in musical diversity. In fact many people expressed interest in CompMusic and on the problems we have started to work on. I hope this is just the beginning and that ISMIR 2012 in Porto will have a stronger presence of papers working wth a wider variety of music repertoires and on methodologies that might be applied to culture specific problems. From CompMusic we will do our best to contribute to that.