Recent Entries
- Achieving Enterprise Adoption of Social Media Tools
- The past, present, and future of location-based mobile social networking
- The steady shift to an RSS-based universe
- Peak of Australian Twitter use was at Future of Media Summit 2008
- Media is becoming everything
- How smaller countries and regions can develop their film and screen industries – 5 key issues
- ABC Interview: Google as an advertising aggregator
- Media Trends + Strategy: The State of Play
- Media and Social Networks Roundtable
- How will news and social networks be integrated?
« Previous Entry | Future of Media Summit Blog | Next Entry »
Future of Media Summit Blog
Digital Music: Managing Download Overload
Anne-Marie Roussel, who will be speaking on the San Francisco side of the Future of Media Summit, has a great blog. One of her recent extremely interesting blog posts was on collaborative filtering, a topic I believe will be central to the Future of Media.
The original blog post is here - full text below.
Digital Music: Managing Download Overload
The recent acquisition of the music recommendation site Last.fm by CBS served as another indicator of the explosive growth of the digital music market. Analysts and record companies estimate that digital music sales will multiply 20-fold by 2010.
That’s significant growth –driven by four key factors:
o Large increase in the number of music players sold and the wider device choice for consumers;
o Increased penetration by music-enabled mobile phones
The importance of social networking in shaping demand;
The ever-increasing number of online music retailers: Currently, iTunes commands 90 percent of that market. Today, it is a numbers game – the more content a site offers, the more consumers it’ll draw.
BUT, the digital music market is at a turning point
Tomorrow, it will no longer be a matter of quantity, but quality – ie, the value offered to consumers on top on plain content. The winner in the digital music war won’t be the one who can simply offer the largest music catalog. It will be the one who can help consumers manage the thousands of songs on their playlist. For music device/service providers, best way to capitalize on this digital music explosion will be to innovate – or acquire innovation – in the tools that enable consumer to 1) pre-select what they download and 2) manage the thousands of songs in their music collection (playlist). According to iTunes Registry, the average iPod user has 3,542 songs in his /her collection and actively listens to only 23 percent of them. Sixty-four percent of the songs are never played.
Notable new entrants
Several companies have developed music recognition and recommendation engines that aim to do that – by identifying a song’s musical attributes and then matching it to a listener’s taste. However, today, no software exists in the world that can identify a song by its musical attributes as well as humans. Much innovation is still needed to achieve that goal – thus the market opportunity.
Several companies are developing technologies to address at least part of the issue. Some, like Pandora (the most famous) and Gracenote, have been around for several years. Pandora evaluates songs according to hundreds of attributes (such as melody, harmony and rhythm) with humans (expert musicians) listening to the songs and performing “manual annotation” – ie manually scoring them for each attribute. Pandora’s shortcomings are 1) it focuses only on popular songs thus limiting the music that can be recommended; 2) Speed and efficiency of scoring is limited by human capacity.
Others appeared on the scene only recently - or are still in the R&D stage, like the Music Mood Wheel project by Microsoft Research. There are basically thrtee categories of providers: Those who do 1) Music recognition only; 2) music recommendation only; 3) both recognition/recommendation.
Music recognition
The company that will achieve Music recognition on the most reliable basis will be the market leader. Music recognition is the most difficult function to perform for a piece of software – because it needs to replicate human experience and/or human musical ear. Humans recognize a song if 1) they already know it (experience); 2) they don’t know it but they recognize the artist (experience); 3) they don’t know it but they can recognize the general style of the music and attribute it on the fly to an artist or a genre (musical ear).
The market opportunity for reliable music recognition is that, once the “fingerprints” of songs are identified correctly, it becomes easier to push to consumers music that correspond to their taste (or their mood) based on the musical attributes of the songs they want to hear – and thus to offer the highest value to consumers.
For example, MSR’s Music Mood Wheel technology identifies and classifies music according to two parameters – melody and rhythm of songs - to find the songs that best fit a listener’s given mood (mellow, upbeat, melancholic, happy etc..). Their concept is that when we listen to music, we usually want to create a specific ambiance (ie, party), or reinforce an emotion(ie, happy or sad), or suppress boredom (ie, while driving in the car). With the size of our music collections now exceeding our ability to recall every song, Music Mood Wheel replaces the "recall and search by title" process by a “browsing music by desired emotions” feature. The mobile device they have developed has a screen which listeners navigate via a “wheel” (similar to the old-fashioned radio wheel); the screen shows a graphical representation of music pieces - each song is a dot on the screen. The dots are arranged along to 2 axes: the vertical axis defines melody (the higher vertically, the more melodious, ie piano sonata).); the horizontal axis is the rhythm (the further to the right, the more upbeat the rhythm, ie rap). Listeners click on whatever part of the screen they are in the mood for – say, upper right hand corner for highly melodious and very upbeat - and the device automatically plays songs classified as such.
Music recommendation
This is a different model than recognition. These engines rely mostly on the “wisdom of crowds” to push the right song to the right consumer. These engines that do only music recommendation offer some value to consumers – but in the long run, their performance will pale compared to the ones based on reliable/algorithmic identification of musical attributes. They follow the model introduced by Amazon of a “collaborative filtering” process : “people who purchased this song also liked this other song” (SoundFlavor, MyStrands, BiggerBoat). Or a “community recommendation” model (Last.fm, Shazam), whereby the song of a given genre that is most listened to by the community is the one that will be recommended to a consumer looking for that genre. We expect that the later approach will have some traction in the marketplace –not so much because of its recommendation characteristics , which are not the most reliable, but because it addresses the importance of social networking in shaping demand.