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

If you live in the Princeton NJ area, be sure to head to the Taplin Auditorium tonight (Sunday January 22) for the world premier of PLOrk, the Princeton Laptop Orchestra. PLOrk consists of about 15 or so computer musicians with laptops that perform live music.  Each musician in PLOrk is outfitted with a rather hefty set of hardware that becomes their musical instrument.  This 'instrument' includes music synth hardware, amplifiers, speakers, sensors (tip, tilt, pressure, force, light and IR), keyboard controllers, microphones, along with a powerbook. (Checkout all of the hardware details).  For software each instrument is outfitted with max/msp ahd ChucK. The cool think about ChucK is that it encourages 'performance coding', writing code while you are performing music. What Fun! With ChucK, we programmers can all be rockstars.


 

The Simac project has a pretty good demo of a music recommendation system that uses acoustic similarity to help generate playlists.  (Registration is required, but worth it if you are interested in music similarity). Check it out at: MusicSurfer.  (It seems to have a bit of trouble with Firefox though ...)

Be sure to check out SunFleet, a new blog authored by the folks in Sun Labs that are researching Design and User Experience.  They will be blogging on all sorts of interesting topics including User Experience, Gaming, Agents, Video Collaboration, Social Networks, Media and Entertainment.   Neat post today on Processing a great language for doing interesting visualizations.  Lots of neat videos.  My favorite is Flight Patterns, a visualization of FAA data.

Now from the state of Maine is  Goombah ...  yet another music discovery service.  Goombah relies on social data to make music recommendations. Goombah looks at your iTunes music data and finds other simiilar music listeners and makes recommendations for you based on the listening habits of these similar music listeners.  Yet another variation of the 'people who liked XXX also liked YYY.  Goombah makes its money by referring listeners to music sellers like Amazon and iTunes.

It seems like the social discovery space is getting pretty full. There's last.fm, the itunes registery, MusicMobs, MusicStrands, tons of playlist sites, and of course all the big music sites like iTunes and Yahoo that rely on social data to recommend music.  At this point it seems that any social site that is trying to hook listeners up with new music has to find a way to let the listener actually listen to the music, not just give them 30 second samples.   The 30 second samples are far too short for me to decide if I want to buy a song or not.  This is a tough nut to crack with the current music laws. A service has to either own the rights to the music (like Yahoo or iTunes), have a license to stream the audio (like last.fm), or have some otherway of resolving the music to a music service (like MusicMobs).  Without the music I don't think a recommender can be successful.  Goombah says that they'll be offering some music downloads in a few months so perhaps they'll find a way to make it work.

This week on the Inside the net podcast is an interview with Tim Westergren of Pandora.  Pandora is a content-based music recommender/music streamer.  In the interview, Tim enthusiastically describes how the Pandora and the underlying music genome project works.  According to TIm, at Pandora, they have about 40 musicians that spend their days labelling songs with about 400 different attributes.  It takes about 20 to 30 minutes for their trained musician to rate a song, and they are currently adding about 7,000  songs a month. After six years, Pandora has amassed labels for hundreds of thousands of songs and can do a find job of recommending music based upon similarity.  However, they do have difficulties with scale.  They are only able to add a very small fraction of new music each year and have to pick and choose which songs get added to their database.  They skip some genres completely. For instance, there is no Classical music in Pandora.  One of the advantages of a content-based recommender like Pandora has over the more traditional collaborative filtering models used in systems like last.fm is that they are immune to the popularity bias that is found in the collaborative filtering systems.  A content-based system is just as likely to recommend a garage band as it is to recommend the Beatles, since it is immune to popularity, whereas a collaborative filtering system that is based upon user listening patterns is much more likely to ignore the unknown bands. They don't recommend unknown bands because no one is listening to them.  Content-based recommenders can push listeners into the long tail of music, finding unknowns that sound like music we already like.  Unfortunately for Pandora, the scaling problem makes them less likely to be able push people into the long tail. Pandora has to pick and chose which songs to process, and they will start with the most popular songs first, so they really can't push people too far into the long tail (yet).  Despite these issues, Pandora is really good way to explore and discover new music.  It is worth trying, and since it relies on  flash, it runs on just about any platform out there.. 

At CES this  week, Gracenote announced a new product called 'Discover' that will provide music recommendation services for online music services. Gracenote Discover is using 3 techniques:

• Music experts label the music with 'micro-genres' a fine grained set of genre labels (they claim to use 1600 genres).
• Content-based - using 'DSP technology'
• Community Preferences - collaborative filtering such as one finds at Amazon.com.

Gracenote says that their engine is the Industry's First Global Multi-Method Music Recommendation Engine that "ensures the optimal music is recommended every time". 

The claim 'ensures he optimal music is recommended every time" makes me smile a bit.  Anyone who has worked with any kind of recommender engine, and especially with music recommendation based upon acoustic similarity knows that there is no such thing as an optimal music recommender.  In fact, music information retrieval researchers are still trying to figure out how to compare similarity engines to determine if one is better than another.  The problem, of course,  is that music similarity is a very subjective metric.  I may think two songs are similar because of the lyric content, while someone else may think they are very different based upon the intstrumetation.  There's no right answer.  The same holds true for music recommendation in general. There's never a 'right answer', so saying that a system ensures the optimal music is recommended is, well, rather funny.

Here is the press release: Gracenote Introduces Discover -- Industry's First Global Multi-Method Music Recommendation Engine

During the week long break (At Sun it's a week long holiday between Christmas and New Years), we have lots of time at home to play games.   At the Lamere household we have some rather stringent gaming requirements: the game must allow for 5 or more players (4 kids, mom's not a game player), age range 10,14,15,16 and 45, so we have to avoid the extremely complicated rules, need a 1 hour playtime, and have a high entertainment factor.  This year's game of the year by far (well, we never got to the other games, we've just played this one, we've had so much fun) is Killer Bunnies and the Quest for the magic carrot.  This is a card game where you have to acquire the magic carrot, while keeping your bunnies alive (and trying to destroy your enemy's bunnies).  It's a lot of fun.   The best part of the game is the player interaction. You can threaten, bribe, weedle and  whine your way through the game. Making deals and finding leverage is the only way to win.  You also find out interesting things about each other. For instance, I found out that my 15 year old daughter would rather sacrifice all of her and her siblings bunnies to a nuclear strike (and virtually eliminating any chance of victory for any of them) rather than give in to blackmail.  Woah!

There's a Crazy Apple rumor about a new class action lawsuit against Apple, claiming (with tongue firmly in cheek) that severe contrasts in music styles found in iPod shuffle playlists are causing whiplash in iPod  users, once again proving the need for intelligent automatic playlist generation to  eliminate the shock that can happen when shuffle play puts Rammstein right after Norah Jones.
Thanks Jeff!

Google has just added a new 'music search feature'. For instance, now if you type in 'the Beatles' into the search box, your result page will have an entry at the top that takes you to an artist page that shows the discography, song information, reviews and links to lyrics and online music stores where you can buy the music.  Pretty neat stuff.  First seen by splog.  Also stories at SearchWatch and at Reuters.

Ever since the invention of the mix tape,  sharing of music has been one of the best ways for people to find out about new music.   The original Napster brought this sharing to a new level, where one could browse the music collection of any one of millions of people.  If you found someone in the Napster network that had similar tastes in music, it was easy to look through their entire music collection and see what they may be listening to that you didn't know about. It was a great way to explore new artists.  Now of course, this avenue of music discovery has been shutdown, but there are a number of new services out there that are helping to fill this social gap left by the shutdown. Sites like last.fm, SoundFlavor, MusicStrands,  and the iTunesRegistry, provide ways for people to explore new music by finding others that have similar music tastes.  A new report: Consumer Taste Sharing is Driving the Online Music Business and Democratizing Culture, (first seen on the Digital Music WebLog), provides lots of data to back up the assertion that music sharing is very important to the music industry.  Authors Mike McGuire and Derek Slater suggest that online music services will be more desireable if they provide better ways for users to share music (or at least share their taste in music) with others.  They predict that in a few years, 25% of all online music sales will be as a result of people sharing their music tastes with each other, but before this can happen, copyright holders will need to provide more flexible DRM and licensing rules.

In The Beginning of the End for DRM (via digg) Fred von Lohmann of the EFF asserts that Rhapsody's recent forays into Linux and Mac music distribution is a sign that DRM is on the way out.  Indeed,  I am waiting for the day when Music Flows Like Water, but I don't think that the recording industry is ready to switch models, it seem that they'd much rather see Subpoenas Flow Like Water.

Amazon is always doing cool things with their data. One thing that they've added recently are some text stats. They have the usual readability indices (Fog, Flesch, Flesch-Kincaid), complexity measures  and such.  For instance, here's some data for the classic book Pattern Classification. One 'fun stat' that they have is  words per dollar and words per ounce.

Here's some words per dollar for some of my favorite books:

• Cryptonomicon: 50,053  words/dollar
  
• Quincunx: 33,391 words/dollar
• Good Omens:  15,149 words/dollar
• Goodnight Moon: 16 words/dollar
• Good Dog, Carl:  4 words/dollar

Clearly there's a wide range of words per dollar here, if we somehow  could get Neal Stephenson to write a book in the Good Dog Carl series, I think we'd be able to sell it for about $100,000 a copy , (or maybe it would be 3 for a penny) .... woah!

When I was a teenager, the best way to find out about new music was listening to FM radio.  In the Boston area we could listen to the big mattress and other album oriented rock stations and always find new and interesting music to listen to.  Today however, the FM dial is now a great wasteland if you are interested in new music.  Try this experiment.  Grab a pencil, a piece of paper, sit in your car and scan the FM dial and keep track of how many stations you actually hear music that was recorded in the last 10 years.  It is really quite depressing. Here are my stats (data collected in Nashua NH at 7:24 AM on December 2, 2005).

	Type	  	Station count	Percent
	-------------------------------------------
	Talk Radio	11		37%
	Commercials	 9		30%
	Oldies		 7		23%
	Christmas Music	 2		 7%
	New Music	 1		 3%

That's 1 station out of 30 playing new music during prime commuting time.  I wonder how this could have happened.

The proceedings for ISMIR 2005 have been posted, so if you didn't make it to this year's Music Information Retrieval conference you can do the next best thing and read all the papers. There's quite a range of topics including user studies, audio classification, symbolic classification, toolsets.  A good place to start is A Survey of Music Information Retrieval Systems a survey paper describing the various content-based music information retrieval systems. 

For the hard-core math weenies out there, the paper that seems to have the highest density of equations to text is
the paper Harmonic-Temporal Clustering via Deterministic Annealing EM Algorithm for Audio Feature Extraction by Hirokazu Kameoka and friends with 35 labeled equations (one equation consumes nearly half a page of text).

The paper by Charles Parker called Applications of Binary Classification and Adaptive Boosting to the Query-By-Humming Problem is a quite well written paper describing methods of symbolic alignment using binary classifiers. (I tend to favor papers that rely on pseudocode to explain things as opposed to maths, which is Mr. Parker's approach).

 Foafing the Music: A Music Recommendation System based on RSS Feeds and User Preferences is a MIR meets Web 2.0 paper with lots of good ideas about combining the various source of music information such as Audioscrobbler or MusicStrands.

There are about 700 pages of interesting reading on the ismir 2005 site. If you read 5 papers a week you'll be just about done when ISMIR 2006 rolls around. So get cracking.

From the Onion ... snicker ...  but not too farfetched since the RIAA is going after mashups, satellite radio, college students, P2P, Moms, Grokster,  the analog hole, Europe,  and spending lots of energy justifying rootkits.  But still, there are those who are standing up for our rights.