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

If you live within a day's drive of Montreal, consider going to the Montreal Music and Machine Learning Workshop. The workshop is being held at the Université de Montréal on November 14 and is being organized by Douglas Eck, Michael Mandel, and Pascal Vincent. The schedule will include a small number of long-format talks, a poster session, and plenty of time for informal discussion between machine learning and music researchers. You need not be a machine learning expert to attend!

(And while you are attending, be sure to get Doug to show you their quarter million dollar piano!)

Here's a freakomendation from Amazon. If you like the Madchester band Stone Roses you may like marble flowers.

Halloween is coming, time to find a costume. If you go to Amazon and search for "terrorist costume", Amazon can't find anything, but they do suggest that you might try the related search 'arab costume' - suggesting that 'terrorist' is synonymous with 'arab'.

Of course, this isn't really the view of Amazon. They don't equate terrorists and Arabs. It just that they have an algorithm that looks for statistical patterns in searches that they use to suggest alternate searches. This algorithm notices that the same people who search for 'terrorist costume' will often search for 'arab costume'. This related search that seems to equate terrorists with Arabs is just a reflection of society's prejudices.

This isn't the first time we've seen a racist recommender. Wal*Mart got into a little bit of trouble when their recommender started to associate films about black historic figures like MLK with the movie Planet of the Apes. Since automatic recommenders are just reflections of the biases of the population at large, the recommenders become a mirror of our society. If the recommender is giving racist recommendations, it is likely that the racism exists in the population at large.

A few years ago, for fun, I wrote a mashup web app called SnappRadio that would select and display images from Flickr related to the music you were listening to (on RadioParadise or Last.fm). My buddy, Sten, has just coded up a version of SnappRadio using JavaFX.

SnappRadio FX takes SnappRadio out of the browser and onto the desktop. Sten is using lots of the cool features that are built into JavaFX such as timelines, keyframes, reflection. Interestingly, although SnappRadio FX looks like a full 3D app, Sten says it really is a a 2D app - apparently JavaFX 3D doesn't exist yet. Sten also points out that the JavaFX performance is rather unpredictable: "Four out of five runs of the app may be smooth sailing, and then I’ll get a run with two second pauses between frames. Whaa? I hope this is a by-product of the preview build and that 1.0 will be more consistent."

Sten says that he'll be posting the source code for SnappRadio FX in the future. I'm really interested in seeing how the JavaFX code looks.

You can read more about SnappRadio FX, (as well as give it a try) at Sten's blog. (And while you are there, you might as well add Sten's blog to your regular reading list - he's been writing some great stuff lately).

Pandora has updated their iPhone app - and it really shows. I've been listening to Pandora on the iPhone over the last few days using nothing but the slow iPhone 1.0 edge network - and so far it works seemlessly - no stutters or pauses (except for an initial buffering of the very first song on a stations).

Pandora has done an excellent job at bringing their simple, one-button personalized radio to the iPhone. Check out Pandora's demo video to see how it works.

Last.fm has added song lyrics to their site. You can now search for music based upon lyrics. For example, lets find some songs about ThatOne:

This gives lots of results:

When visiting the song page at Last.fm you are now given an excerpt of the lyrics, with a link to the full lyrics.

Here's the lyric detail page:

The lyrics are a result of a partnership between Last.fm and LyricFind. The press release says over 800,000 lyrics have been included.

Last.fm has done a nice job of integrating the lyrics into the website, but it would be nice if they could get them integrated into their standalone player. It would also be really neat if they could start to include lyric similarity into their computation of artist and track similarity. I'd like to be able to build playlists and radio stations based upon lyric similarity. (I did some experiments along these lines a while back with fun results).

First, the good news: the best online streaming music app - bar none - is launching today.

Spotify has just announced music licensing deals with Universal, Sony, BMG, EMI Music, Warner Music, Merlin, The Orchard and Bonnier Amigo. With those deals in place, Spotify has gone live. Premium access (advertising-free, paid subscription) is available now (October 7), while free, ad-supported access will be gradually made available over the next few months.

Now the bad news: Unfortunately, for those of us in the states, Spotify is only launching in the UK, Germany, France, Italy, Spain, Finland, Norway and Sweden. No word yet on when they will be launching here.

Spotify is a streaming internet music application that offers on-demand access to a seemingly unlimited amount of music. In some ways, it is like iTunes - you can search and browse for music, create playlists, and play music by any artist with just a click of the mouse. But Spotify is so much better than iTunes since it gives you access to millions of tracks (all perfectly legal). With Spotify you can share playlists with other Spotify users (and they can actually listen to them), you can even create collaborative playlists where multiple users can add tracks to the list. Spotify also offers a Pandora-like radio mode that will give you an infinite playlist of music by similar artists (using AllMusic's artist similarity). You might expect that since Spotify is streaming music over the wire, that it would be slow and cumbersome - but to me, it seems to be as fast or faster than iTunes. When I click on a song, it starts playing immediately - with no stutters or drop outs (well except for that time when my wireless router was dying).

In addition to the excellent music client, there's a lot of cool stuff going on in the Spotify back room as well. Every item (artists, tracks, albums and playlists) in Spotify has a URL associated with it. This makes it easy not only to share music but for third parties to build applications that use Spotify as the music engine. (We've done that here in the Labs with our Music Explaura - when you click on the little green play button you are playing music in Spotify). Spotify also provides web services to allow 3rd parties the ability to search the Spotify catalog for artists and tracks and resolve them to Spotify URLs.

The premium subscription costs £10 per month or £100 per year in the UK. It comes with these attributes:

(Note that the premium subscription comes with two beta invites, so if you live in the USA, it is time to make a European friend). Spotify is also offering a day pass - (£1 per day) for when you are planning that romantic dinner and you don't have enough Barry White to get you through the night.

I've written about Spotify a few times already. Since then, the Spotify team has added lots of new features. For instance, there's now a 'What's New' tab that highlights music that has been recently added to the catalog, along with some artist recommendations:

They've also added a zeitgeist page that shows you what is popular on Spotify (or just what's been popular for you).

Spotify is the closest thing there is to the celestial jukebox - with (nearly) all music available on demand. I'm really excited to see where the Spotify team takes this. Congrats to the Spotify team for releasing such a great music app - (but don't spend too much time resting on your laurels - get back to work on those pesky music deals so that Spotify can be released in the U.S.!)

I was looking through the list of proposed panels for the next SXSW to see what talks and panels I could find about music recommendation. I found a number of interesting talks that were related to music recommendation, but none that seemed to answer the question about why music recommendation seems so broken and what is going to be done to fix it. Since I know a bit about the topic, I decided that perhaps I should propose such a talk. But I found out that I'm way too late ... the SXSW proposal deadline was July 11. However, the SXSW organizer that I contacted said that there is still a very small chance that if I submitted something that it could be accepted. He indicated that although the voting on submissions is closed, that if my proposal received some favorable comments, that would help. So although it is a long shot, I decided it wouldn't hurt to submit something, so here's my proposal:

I'm so sad, my iPod thinks I'm Emo. - Music recommendation is broken - automatic music recommenders make mistakes that no human would ever make. In this talk, we will explore why recommenders make such dumb mistakes and we will explore some of the new ideas coming from recommendation and music researchers to help make music recommendations better.

If you think it is interesting and may be something you'd attend at SXSW, I'd be pleased if you'd add a comment to the proposal.

Here is a list of other SXSW panel proposals that are somewhat related to either music discovery or recommendation. It looks to me like the most closely related panel is Music 2.0 = Music Discovery Chaos?. This one does look interesting and I'd love to attend it, although I don't think it is going to be too research-oriented.

• I'm so sad, my iPod thinks I'm Emo. - Music recommendation is broken - automatic music recommenders make mistakes that no human would ever make. In this talk, we will explore why recommenders make such dumb mistakes and we will explore some of the new ideas coming from recommendation and music researchers to help make music recommendations better.
• Keeping it Human in the Age of Big Data"The selection and placement of stories on this page were determined automatically by a computer program." - http://news.google.com Or was it? At Last.fm we've grappled with how to present and filter our user-generated music data, but also how to mix in true editorial content. Humanizing core features was much of the impetus for the recent re-launch.
• Music 2.0 = Music Discovery Chaos?The way we discover music has entirely changed in less than 10 years. Radio’s aging demo is presented with safe mainstream offerings. Music discovery is at the forefront of technology and social networks, yet no new standard has successfully been adopted. Websites abound attempt at both data and user generated rating/filter systems. Human VS algorithm: what method can save us?
• Neither Fish Nor Flesh: Music Discovery Going Semantic - Current music discovery techniques rely on human annotation and human behaviour to generate recommendations. We review current recommendation methods and take a in-depth look at new semantic recommendation methods, and discuss how both worlds can be combined to deliver an ultimately better music discovery experience.
• “Just For You”… Really? - Personalization technologies are all over the web now, with applications everywhere offering recommendations for movies, music, video, blogs and most anything else we might like online. Do any of these systems work, and if so how? Is “Discovery” the next “Search,” or just an excuse for behavioral targeting systems that compromise privacy? Finally… If you’re a site publisher interested in adding personalized recommendations, how do you pick a partner to help you get there, or strike out on your own and get the feature built????
• Collaborative Filters: The Evolution of Recommendation Engines - Recommendation engines have boomed in the era of social media. These panelists are experts at collaborative filtering systems. Citing Digg, Amazon, and Netflix as examples, they will have a high level discussion about the evolution of recommendation engines and how each approach is different.
• People Who Purchased This Also Purchased… What?? - Ever wonder how sites determine what music and movies to recommend? Ever received an off-beat recommendation that made you go hmm….? That’s because most sites use collaborative filtering – which determines recommendations based on what other users like. This panel will discuss how these technologies are evolving and the future of recommendations.
• Algorithms, Meta-Data, and Why the Future is Behavioral - Data about data or data with data? Social smarts versus machine smarts? Recommendation engines are finally delivering on helping us filter infinite choices while user generated ratings and reviews can now be found on almost any product or service. Where are these being combined to help users and what's next?

Lookie here ... it's a Last.fm commercial - first time I've seen one of these ...

The Screw Up - Last.FM Commercial
- I get the joke, but I must admit that I'm not too crazy about this one (perhaps because I have 3 teenage daughters...). Well, maybe this is just a video made by a Last.fm fan and not a real commercial.

In response to Pandora's recent warnings that it may go out of business due to the ongoing music licensing issues, nuTsie (yet another music streaming service) has issued A nuTsie challenge:

The Challenge dares every Pandora user to try nuTsie for one day and then to let us know which service they prefer. If Pandora has more votes than nuTsie at the end of this October, nuTsie's own Dave Dederer, founding member of the rock band The Presidents of the United States of America, will write and record a witheringly sincere love song to Pandora's CEO, Tim Westergren, and put it on the nuTsie.com home page for a week. Dave invites Tim to reciprocate the bet.

The folks at nuTsie do have a sense of humor: The nuTsie Challenge is a creation of nuTsie's elite in-house think tank, the Music Gnome Project. The Music Gnome Project believes that music is magical and strange and unpredictable, not some geeky science experiment.

However, I think Dave had better start tuning his guitar ... my nuTsie trial didn't go too well. I started with alternative metal band 'Breaking Benjamin'

nutSie was able to play a few tracks by Breaking Benjamin, but then took a hard left turn and started to play the Simple Symphony, by Benjamin Britten. Ouch! This was followed shortly thereafter by South African jazz vocalist Sathima Bea Benjamin. The nuTsie song transitions were giving me iPod whiplash as I was dragged through a number of wildy different genres.

.

It looks like the Music Gnome project is having a bit of a hard time distinguishing between all of the different artists with 'benjamin' in their names. That certainly would explain why they think that music is magical and strange and unpredictable.

I'll give nuTsie a fair trial for a whole day, but I suspect if nuTsie continue to make these very basic metatdata mistakes that Tim and the rest of the Pandora gang shouldn't be too worried - I suspect that in the war between the Gnomes and the Geeks, the Geeks will win.

Netflix has just released a new developer API. The API gives access to metadata for 100,000 movie titles and also gives access to user data, allows 3rd parties to manipulate user rental queues and get recommendations. The API uses OAuth for authentication. Perhaps one of the most interesting thing about the API is that they allow for commercial use. Cool stuff.

At ISMIR, I asked Anthony Volodkin, founder of the Hype Machine what music sites he was excited about. He pointed me to thesixtyone. I've been playing with thesixtyone for about a week now and Anthony is right, thesixtyone is really interesting.

The sixtyone is part music discovery site, part social music site, and part online game. On thesixtyone, you listen to music (almost all of it by artists you've never heard of). If you find something you like you can 'bump it' - (it's just like digging) - songs or artists that get bumped a lot hit the home page, where they get lots of visibility and lots of plays (just like Digg). Not only is getting to the front page highly rewarding for the artist, it also can be highly rewarding for you if you were an early 'bumper' of the track. That's because everything that you do on the site can earn you points. Visit the site every day, you get points. Listen to music from 'the rack' (the new bin), you get points, bump a song that then gets bumped by lots of others, you earn lots of points. Be the first person to bump a song that ultimately makes it to the front page - you get a boatload of points. As you earn points, you go up levels, just like D&D - and the higher your level, the more weight your opinion carries - higher level users can bump songs multiple times. You also get points for 'achievements' like 'disk jockey' if you attract lots of others to your 'radio station'.

I've really just scratched the surface - there is lots to do at thesixtyone - you can spend time looking for undiscovered gems, you can listen to the music that is been bumped the most, you can explore the 'favorites' of other users, you can build playlists, you can create your radio station, you can comment on tracks (the comments get turned into little popups that sometimes appear when people listen to the song) - it really does feel a bit like a D&D game - it is fun to notified via a little popup that you've just received some points or achieved some goal. And while you are doing this, you are listening to music, of all sorts of genres. The music is mostly good to excellent (with some clunkers too). And there are some big name artists - I've listened to NiN, Bjork, Jonathan Coulton and Daft Punk on thesixtyone.

thesixtyone developers have done a great job of making the site highly interactive. It is very highly polished, and is simply fun to use. The music player is integrated perfectly into the site, so no matter what you do, the music keeps on playing.

thesixtyone is a fantastic place for new artists to get their music into the ears of listeners since listeners are rewarded for finding and bumping the newest music. My friend, Sten, uploaded some tracks for his band "Hungry Fathers" and within minutes their songs had been listened to by half-a-dozen listeners. That just doesn't happen on Myspace.

It has been a long while since I've been excited by a new music site. Most 'new' music sites are just variations on the Last.fm theme - ("Let's build a social community around music!") and there just isn't any need to do that anymore, Last.fm does it so well. But thesixtyone is really doing something different - they are making it easy, fun and rewarding to explore and discover new music.

The observant browser will notice that my link to thesixtyone includes me as a referrer - yes, I'll get thesixtyone points if you follow that link and register

One of the new attendees at this year ISMIR (the Music Information Retrieval conference) was Anthony Volodkin, the founder of the music discovery site The Hype Machine. Anthony is part of the web 2.0 startup culture so he certainly knows about geeks. Here are some of Anthony's tweets during the conference:

• Haven't been so repeatedly and consistently outgeeked this way in a long time, ISMIR rocks! 06:17 PM September 15, 2008
• ISMIR's geeks outgeek Web 2.0 NY geeks by a massive margin 02:21 PM September 15, 2008
• ISMIR 2008 is hardcore-geeky, I love it! First talk is on relationship of geometry and musical consonance... !!! 06:16 AM September 15, 2008

Anthony is a great guy to talk to - he's really smart, understands better than almost anyone what it takes to build a site that people will want to visit, and perhaps most important of all - he is extremely passionate about music - which is key to success in the music 2.0 world.

Doug likes to mock me for my taste for Emerson Lake and Palmer - but I challenge him to watch these videos and not be totally convinced ...

The Flying Piano

Organ Jujitsu

Organ Stabbing

Upside down organ playing

More organ wrestling

(includes some tagging too - along with Toccata and fugue in d minor, while flat on his back upside down).

Hoedown

(with an armadillo like costume)

At the Isle of Wight

Their debut performance - including the post-set fire.

I loved this show when I was 7 years old, Now it is on hulu. Watching it again after 40+ years is really fun, and it is actually much better than I was expecting. With all the Irwin Allen shows on Hulu, who needs 'heroes'?