Duke Listens!: Visit my main blog at MusicMachinery.com

The ISMIR conference organizers have created a virtual poster session in second life.  Point your second life client at: secondlife://Kuusamo/166/88/111  and you will get the chance to browse through all of the posters in the conference.  (If you have a screenshot of the poster session, please send it my way, I don't have a Second Life client).

Update: Adam has posted some images on flickr :

Last month, I wrote about what I thought were the 4 things that Amazon needed to do to be a serious competitor to iTunes.  Well, the beta version of the Amazon.com store is here.  So lets see how they did. 

• Work seamlessly with the iPod  - If I, (or more importantly, a non-digital-music guru), can't get the music from the Amazon store onto their nano or their shuffle, then the store is a non-starter. How they did:  A - they have a tag-along app that will move your Amazon purchases into iTunes for you. (I haven't tried it yet, as I am linux-bound all week, but I hear that it works well)
  
• Broad and Deep Catalog - The Amazon bookstore is a poster child for the long tail - they better do the same for music. How they did  C-  The Amazon store has "over 2 million tracks" which doesn't compare well with iTunes which currently claims 6 million tracks.  
  
• DRM-Free - Rumor is that Amazon is doing the right thing here ... no DRM for their tracks. How they did: A+  DRM-free MP3 tracks  - with 256k vbr encoding  for $.89. Very nice.
  
• Discovery and Recommendation - Amazon is well known for its recommendation and personalization technologies.  If they can make it as easy to find new music as they make it finding new books then Jeff may give Steve a run for his money.  How they did: Incomplete -  they are doing recommendations based on my previous purchases, but when I try to get recommendations for 'new music' or 'soon to be released' music - I'm told that Amazon has no recommendations for me, and they just give me the 'Top Sellers'  - I expect Amazon to do better than that, much better!
  

  One nice surprise: Variable pricing - there are many tracks and albums the sell for lower than the $.89. For instance you can get a DRM-free MP3 version of Pink Floyd's "Wish you were here" for $5.60. That's a great deal for  a super album.

Overall : With their 'beta' label, Amazon is expecting an 'Incomplete' grade here.  If they fill out their catalog and raise the quality of the recommendations to the level of their book recommendations, they should be able to get that incomplete turned into an A.

You can tell these are "Music 2.0" guys and not academics because of their attire and their incredibly serious demeanor.

 

Amazonmp3.com has launched.  You can buy DRM free tracks for $.89, with 2 million to chose from.  Amazon thankfully provides a helper app that will assist in moving the the track into iTunes.    Since Amazon knows about me, it was able to make recommendations right away based on my several hundred previous (mostly non-music) purchases.  The recommendations seemed to be pretty good. I purchased KT Tunstualls 'Suddenly I See" - the track was variable bit encoding with an avg encoding of 256 kbs. 

 Interestingly enough there's a number embedded in the mp3 ID3v1 tag:

Amazon.com: 202109900

 I wonder if this is an ID unique to me, so they can track if/when I share the song with my friends.

Purchasing a track from the Amazon store was incredibly easy. Amazon already has my billing info so it was really a 2 click purchase (I had to click on a terms of use box) - the terms of use tell you that you are not really buying a track, you are just licensing it.

That's my 5 minute look at the Amazon store, so far it looks pretty good.  They really could give iTunes a run for their money.  I know I shall be shopping there for music before I go to iTunes.

Sunday was Day 0 of ISMIR - the tutorial day before the main conference starts. Oscar and I gave our music recommendation tutorial in the afternoon. It was well attended, people seemed to enjoy it and had nice things to say after (but really, does anyone ever have anything bad to say?). I would have liked to attend some of the other tutorials, but I was too busy prepping for our tutorial.

At the end of the day they had a wine reception with oodles of local wines to sample. It was a great introduction to the wine culture in Vienna.

Some folks are already starting to post images to flickr. Checkout the ISMIR 2007 gallery.

Last night, a bunch of us when to the Schnitzelwirt for dinner. Good food, great conversations. Here's Jeremy, Masataka and myself before the veritable mountain of schnitzel arrived.

After a nap to make up for a sleepless redeye, I took a stroll through Vienna. I headed over to the TUV, where the conference is hosted - no one was there ... so I just wandered from there up the main shopping street to St. Stephens. There was a clown festival, I've seen more than my share of adult men with red balls on their noses today. After the walk, I figured out how to use the subway and made it back to the hotel without getting lost. Not to shabby. In a few minutes I'll be back out again heading to dinner - then (hopefully) a good night sleep, since I have a lot of talking to do tomorrow at the music recommendation tutorial.

I had to make sure that the camera/laptop/flickr connection was working before I hop on the plane to Vienna, so I took this picture of some of my favorite MIR-related books. I haven't read 'The Influentials' yet, but it does look interesting and relevant to music recommendations, tastemakers and trendsetters.

Oh, and if you are taking pictures at ISMIR, be sure to upload them to Flickr and tag them with ismir2007, so we can have a shared photo gallery. (There are about 100 shots on flickr from ISMIR 2006).

I think that we will see a big shift in ISMIR this year to a focus on the social side of music, music retrieval, social tags and recommendation. Looking through the list of papers and posters, there seem to be lots that are related to social music, including:

• Exploring Mood Metadata: Relationships with Genre, Artist and Usage Metadata
• Virtual Communities for Creating Shared Music Channels
• jWebMiner: A Web-Based Feature Extractor
• Meaningfully Browsing Music Services
  
• Web-Based Detection of Music Band Members and Line-Up
• Tool Play Live: Dealing with Ambiguity in Artist Similarity Mining from the Web
•  Music Recommendation Mapping and Interface Based on Structural Network Entropy
•  Sociology and Music Recommendation Systems
•  Creating a Simplified Music Mood Classification Ground-Truth Set
•  Preliminary Analyses of Information Features Provided by Users for Identifying Music
•  TagATune: A Game for Music and Sound Annotation
• A Web-Based Game for Collecting Music Metadata
  
• Autotagging Music Using Supervised Machine Learning
• A Semantic Space for Music Derived from Social Tags 
• Similarity Based on Rating Data
• The Quest for Ground Truth in Musical Artist Tagging in the Social Web Era
• Annotating Music Collections: How Content-Based Similarity Helps to Propagate Labels
• A Game-Based Approach for Collecting Semantic Annotations of Music
• Human Similarity Judgments: Implications for the Design of Formal Evaluations

 

It is almost time to head off to Vienna for ISMIR, the 8th International Conference on Music Information Retrieval. Sun joins a number of other companies including Phillips, Bang& Olufsen and Music IP in sponsoring this conference.  If you are attending the conference be sure to checkout some of the MIR work that we are doing at Sun labs.  Thierry Bertin-Mahieux from the University of Montreal is presenting some of the work that he, Doug Eck and I have been working on: Autotagging Music Using Supervised Machine Learning.  Also I'll be showing some of the visualizations we've been using to help people explore their music spaces: Using 3D Visualizations to Explore and Discover Music .  And in just a few days, I'll be presenting the Music Recommendation Tutorial with Oscar Celma.  I'm really looking forward to this, I think it should be a fun talk - and it has been just great working with Oscar on this.

Here's a strange iTunes recommendation. Based on the items in the shopping car: 'Selected Chants of the Russian Orthodox Church' and 'A Gregorian Chant' by Benedictine Monks, iTunes is recommending Green Day, Interpol, The Mars Volta, Velvet Revolver Joni Mitchell, and Amy Winehouse. Me thinks they are having a bit of a sparse data issue here. Not enough people have listened to Gregorian chants to allow iTunes to tap into the wisdom of the crowds.

Or is this a veiled reference to Iran? A recommendation from iTunes

On the MIR Research blog, Elias points out that the team that organizes MIREX scored highest in a number of categories. Elias suggests that having insider knowledge of the specifics of an evaluation can give a submission an unfair advantage.   I can see how this can happen, even unintentionally.  A system that is built and tuned against a particular collection of music could be overfitting the collection.  Now if you bring this overfitting system to a new collection, its performance will tend to drop since it is not tuned to for the new collection. The overfitting hurts your system.  However, if you are testing, training and tuning against a test collection that happens to overlap the evaluation collection,  then it is possible that overfitting can help your system in the final evaluation.  So for example, if the IMIRSEL system happened to be built and tuned against the same collection that was used in the MIREX evaluations, this may give that system an advantage. Knowing how difficult it is in this day of copyright litigation to build up a sizable test collection, it is not hard to imagine the IMIRSEL team using the MIREX evaluation collection to tune their system.

Now, I have worked closely with just about all of the members of the IMIRSEL team, and I know that they are an extremely talented set of individuals and are quite capable of creating a winning system.  I also know that these folks are diligent, methodical and careful in how they design the MIREX evaluation to ensure the utmost in fairness.  I have little doubt that the MIREX results accurately reflect the state-of-the-art.  However, not everyone has had the opportunity of working with the IMIRSEL team and so not everyone may share the same confidence in the MIREX results.  To alleviate these concerns, I suggest that the IMIRSEL team detail the datasets used in training and tuning the IMIRSEL submission, highlighting any overlaps with the data used in the MIREX evaluation.  Such a disclosure will help alleviate any of the concerns some folks may have about the fairness of MIREX.  For MIREX to be a viable long term evaluation, MIR researchers have to have a high confidence that it is fair.

A visit to MusicMobs.com today will bring you to the register.com's domain parking page.  Does this signal the demise of another (and my favorite) playlisting site? Is this just an oversight by Toby? Or is this the first indication of a new chapter in MusicMobs?

Update:  MusicMobs is back.  Perhaps it was just a glitch with the DNS.  Nothing to see here move along.

to sign up for the music recommendation tutorial that Oscar and I are giving at ISMIR next Sunday.  The conference organizers say that they are still taking online registrations, and that on-site registrations will also be accepted.