Read/WriteWeb Files: Online Music
Every week we have a feature called Read/WriteWeb Files, in which we investigate a current hot topic or company in Web technology. This week we're going to focus on Online Music, something that is becoming more and more prevelant as broadband speed increases and social software functionality gets better. Our network blog on digital lifestyle, last100.com, will also be focusing on Online Music this week and AltSearchEngines will list their Top 10 music search engines. So I'm quite excited by what we'll discover over the following 5 days about online music!
When you think of music on the Web, there are roughly three main eras:
1) In the time of the Dot Com companies, it was P2P systems such as Napster and Kazaa that defined online music. Ultimately though P2P systems were defeated by the record companies and their heavy-handed legal tactics.
2) Enter the iPod and iTunes, which define the current era of online music. Apple's combo of a killer devide (iPod) with an online music storage system that syncs with the device (iTunes), has come to dominate the market. You can hardly walk down a city street these days without seeing those familiar white earplugs in someone's ears. Competitors such as Microsoft (Zune) and Real Networks (Rhapsody) are in the game too, but they're a long way behind Apple and likely to remain so.
3) Perhaps the next defining period in music will be streaming music over the Internet. Already this era is underway, with startups like last.fm (recently acquired by CBS) and Pandora making their mark. Most of the big Internet and media companies have horses in this race - Yahoo Music, MSN Music and AOL Music, to name a few.
Beyond iTunes
This new generation of online music has a big focus on recommendations, personalization and social networking. There are a multitude of startups looking to become The Next Big Thing, from Fairtilizer (our review) to SeeqPod (our review) to MyStrands (who recently took $25 M in funding to enhance their personalization system). Even Apple, traditionally not big on social software, has recognized that it can't sit back on its laurels in online music - they recently announced My iTunes widgets, to enable you to share your music, movies and other media with friends.
My iTunes
So this week the Read/WriteWeb Network (R/WW, last100 and AltSearchEngines) will be exploring online music. If you have any suggestions of things to look at, please leave a comment. Also check out this week's R/WW poll: What is your favorite online music streaming service?
Share4 TrackBacks
TrackBack URL for this entry: http://www.readwriteweb.com/cgi-bin/mt/mt-tb.cgi/1530
Comments
Subscribe to comments for this post OR Subscribe to comments for all ReadWriteWeb posts
Richard,
We are just rolling out a new music recommendation and personalization service at ZuKool Music, www.zukool.com
We currently have around 600,000 tracks and will be expanding that as we move forward. We are also in the process of building Facebook/Myspace applications/widgets to allow you to use the service wherever you are.
While we have a user facing site we will be primarily focused on a B2B service (REST based API), providing recommendations for other services (such as muisc portals or e-commerce services).
Its a fun service to play with too.
I would add to the list of vendors which develop state-of-the -art content music recommender systems available today, is BMat from Spain and DoubleV3 from Canada. These 2 vendors use state-of-the-art DSP (digital signal processing) and Machine Learning (ML) algorithms for their products.
Based on what I know in DSP and ML , which are my specialist domains of expertise, these 2 vendors are above all the rests, compared to other content-based music recommender systems.
The future of music recommendation will be content-based rather than annotated text-based or word-based of song items. Content-based retrieval models how human listens to music, in that mean that if you listen to a content of a song, eg: if one uploads a short sample file of Michael Jackson's Thriller song, the engine retrieves all songs (or top n-th songs) that are similar in tunes to the target query song, which is Thriller. This is more accurate than text-based query, which is the most adopted music recommendation engines of today. Amazon still uses user-rating and text-based annotated retrieval for its recommendation. The target or short sample file to be queried doesn't have to be a pre-recorded digital file of the Thriller song, it could be someone humming the tune of Thriller into a microphone which is converted to a digital file, then this file is searched against the data-base for similar tunes. This is more similar to the capability of humans. You can do query by humming using the 2 products mentioned above, however their technologies are still in development to perfect such query by humming, although they have been designed to do query by uploading a pre-recorded sample digital music file.
Ian Wilson,
The peer reviewed research papers (PDF format) from Music Technology Group (MTG), where BMat is a spun-off company are freely downloadable from here. Some of those research publications are used in BMat's product development for their music recommender systems . You might find those papers useful in your product development.
Podcasts are an interesting online music delivery method because of the ability of fans to subscribe via RSS. Artists such as Quincy Jones, Norah Jones, Elvis Costello and Bjork have begun releasing podcasts that include interviews, acoustic versions and video of the artists during the creative and production process. Archived content such as interviews, outakes and backstage footage from many artists including the Beatles has recently begun emerging in the form of podcasts as well.
The portability from desktop to MP3 player and soon via wireless and IPTV makes this a very attractive medium for widespread promotion and content distribution. With the ability to integrate video, audio, presentations and text, podcasts seem poised to emerge as an important promotional tool for musicians utilizing the web.
@falafulu
Thanks for the pointers, appreciated.
There is a lot of good stuff coming out of the Barcelona music technology group, and has been for some time. The MyStrands development team is also based in Barcelona.
Content based systems have many advantages over either meta data based systems or social filtering type systems, however they are not the future as you might imagine. They have their weaknesses too. Most commercial systems I believe will and are moving towards a mix of each of these approaches.
Please check out Jamendo :
Jamendo unlimited, free and legal music download.
Jamendo promotes, filter, distribute Creative Commons licensed music and share revenues with artists.
http://www.jamendo.com
Jamendo is 100% creative commons licensed music, available in P2P protocols at low cost distribution.
Richard,
Interesting topic to pick up for the week and very timely too. Given what's happening with the music corps in the DRM space, it will be interesting to see where things land.
Check out my post from last night discussing Universal/Google/gBox combo and their impact on Apple's revenue stream.
http://abhishek.tiwari.com/2007/08/13/should-apple-fear-the-universal-google-and-gbox-tag-team/
Let me know what you think.
Abhishek
Richard, you may want to consider adding www.algeka.com to your review....we just launched June 15th. At this point we are just taking music videos, but will soon just audios as well.
We have two University professors (with math and computer science doctorates) doing all of the "back end" architecture.
In six months it will have unsigned singers from all over the world. Thanks much, Norm
Hi,
My small contribution, I try to create something (Musiic.net) to federate online music services, you can check it at http://www.musiic.net.
But it is still a alpha version ...
Victor
Thanks all for your excellent comments - lots for us to follow up with here :-)
I'd love to see some insight into The Hype Machine (hypem.com). It's really the only online, streaming site I visit.
You're off to a bad start with incorrect information.
Rhapsody launched at the same time as the first iPods, and has always been a streaming music service, first and foremost. Unlike Pandora, it offers full on-demand listening.
Rhapsody pioneered the subscription "pay once, listen as much as you want" model that Yahoo, Napster 2.0 (itself a repackaging of an older industry-backed service called PressPlay), and Zune have all emulated.
Rhapsody added track purchases later, and only last year started doing tighter integrations with portable devices.
Rhapsody does believe that "streaming music over the Internet" is the future, as you note. That's why they've been leading the way there!
Ian Watson said...
Most commercial systems I believe will and are moving towards a mix of each of these approaches.
May be you're right here, but I don't know how much would such a system depend on content-based and how much on metadata-based. I suspect that it would be towards more on content-based, and this is based on what I read from the literatures .
I have read quite a few of the Music Technology Group papers, since I was going to develop a music recommender system Java API for an e-commerce online shop here in New Zealand, but it was decided that may look at getting an off-the-shelf product.
I am impressed with BMat's product, since they are the only one at the moment in the market that implements BSS (Blind Source Separation) algorithms such as ICA (Independent Component Analysis). BSS (aka - cocktail party problem) models on how human brain manages to filter out different audio components in a noisy environment. Eg, when you are in a cocktail party in a noisy bar , all the audio components arrive at your ears simultaneously, however one still manages to hear whoever he/she is speaking to, since all the mixed audio signals pass thru the ears, the processing takes place in the brain where it separates all the different audio components that originate from different direction in that noisy room (loud music's component from the background, person-A's conversation's component, person-B's conversation's component, a crying baby's component from the background and so forth ). It (brain) identifies, the component that is interested in listening to (the person you're currently talking to), allow this component to pass thru, and thus reject the rest (the uninteresting components). BSS & ICA algorithms separates audio signals into their primary independent components similar to what a prism does to a white light beam, ie, it splits the beam into its 7 primary colors. In a sense, BSS & ICA algorithms are sort of audio-signal prism.
BMat's product works well in noisy environment since the background noise component could be identified and eliminated. So, BMat's product could do query by a single component of the song, eg, the bass-guitar only.
In addition to BSS & ICA algorithms, I believe that they also use Wavelet Transform and Short-time Fourier, for signal decomposition, prior to feed it to a learning algorithm such as Artificial Neural Network for classification.
I do use wavelets & ICA for a financial modeling application, but I haven't used them in any music recommendation engine development and that was the reason, I was exploring to develop one for this local ecommerce shop, since I am already familiar with most of the algorithms that involve in such system.