Squeezing an ISMIR Paper

The last days I've spent fighting Latex. Trying to squeeze as much unnecessary space as possible out of ISMIR papers. (Of course I've also been shortening the contents, but I've reached the point where I really don't feel like deleting/shortening any more paragraphs.) Btw, did you know, that you can fit about 25 references into half a page? Right now I'm debating if a baseline stretch of 0.97 is too low. I can't really tell the difference when looking at a paragraph, but overall the paper is 1/4 page shorter :-)

MIREX 2007 Preparation Meeting

MIREX is all about evaluating MIR research. Evaluations are needed to measure improvements, and thus enable improvements. Even only talking about how to conduct proper evaluations (which is what happens all the time at MIREX) is already beneficial for MIR.

Next week the MIREX 2007 preparation meeting will take place in beautiful Vienna. 25 people from all over the world are on the list of people who will attend the MIREX planing meeting. Every major European MIR lab seems to be sending someone, and there's lots of people from North America, and even from Japan and New Zealand! And I heard that a company working on MIR topics might be sending someone. And of course I'll be there, too :-)

When the evaluations started in 2004 by some fine researchers at MTG no one could have ever foreseen this success. And without Stephen and his IMIRSEL team at UIUC it would never have happened. Even a week ago I never would have thought that Stephen would manage to bring together so many researchers from all over the world for this preparation meeting.

Of course, everyone will now argue that the large number of participants is just a typical Vienna effect (everyone wants the chance to get to see the city, which I fully understand). Btw, as most of us know, this is also the main reason why the number of submitted papers to this year's ISMIR will be much higher than ever before. I hope my papers don't get rejected :-/

Anyway, this year's MIREX success is also based on the success of the previous MIREX evaluations. I know many have been very skeptical about MIREX. But many of the issues we've seen have been fixed, and with some luck this year everything will be really smooth. There's still time to jump on the train! (If you haven't done so already) :-)

Guessing Games

nest @ The Echo Nest is a blog by the Echo Nest guys. There's a post mentioning artist identification results for 2005. With numbers showing that a system named "Cepco" outperforms MFCCs.

I asked Google (maybe I didn't ask correctly) but Google doesn't seem to be aware of any publications mentioning a Cepco system (sounds like CEPstral COefficients, doesn't it?). In 2005 Mandel & Ellis submitted a nice approach to artist identification using MFCCs and SVMs. I'm guessing that at about the same time the Echo Nest found a similar or maybe nicer solution to artist ID?

They also have a post on the world of music criticism. The numbers could be anything (e.g. term frequencies?) but I guess it does mean that Brian and his team have continued research on automatic record reviews?

My favorite is this interesting post. What's so different in the two figures? Anyone have a guess? :-)

If it's a visualization of features extracted from audio then here are some guesses I'd make: Maybe everything (input, algorithms) was the same for both but one of the audio signals was slightly shifted in time? Or maybe they used some rescaling (e.g. loudness compression) techniques on one of them and not on the other? Or maybe one is a cover version of the other?

Recent PhD

Tetsuro Kitahara completed his PhD in March. The title of his thesis is: "Computational Musical Instrument Recognition and its Application to Content-based Music Information Retrieval" (PDF). It's in English! (It's the first Japanese MIR related thesis written in English which I'm aware of.) He wrote it while working at Okuno's Laboratory at Kyoto University. He has now joined CrestMuse (an English version of the project web page is on its way) as postdoc at Kwansei Gakuin University (Katayose's Laboratory). I saw one of his demonstrations last year and was quite impressed.

(The list of MIR related PhD theses and dissertations has been updated.)

Danceability

Last week I was lucky to have the chance to listen to invited talks by Sebastian Streich and Bee Suan Ong (both currently working at Yamaha in Japan). Bee Suan talked about some of her work on the structural analysis of music. I wouldn't be surprised if her approach would have easily scored highest in last year's MIREX cover song identification task. Unfortunately she was too busy to participate (finishing her PhD, getting married, moving to Japan, ...). Also Sebastian's talk was very interesting. Generally, complexity in music is a very interesting topic. He wrote a nice discussion on it in chapter 2 of his thesis. The one thing I found most interesting from his talk was the following figure (page 90 in his thesis):



Basically it shows how his danceability feature that he extracts from audio correlates with descriptors you find on AMG. I find the correlations rather impressive. The curves are very smooth (although he does not display any measure for the variance this indicates that the variance is low enough), and the correlations are meaningful. For example, party music is more danceable than romantic music. I think his graph is a very intuitive way to communicate what his extracted feature does and how it relates to concepts that we use when talking about music. Personally I find it magnitudes more useful than an evaluation showing that if you include a certain feature in a black box genre/mood/style/whatever classifier, then the performance increases by 5%.