All those songs

Although the amount of research around the idea of querying a database of songs by singing or humming a piece of a tune is considerable, is surprising there are no relevant implementations; especially when we consider the success of commercial ventures such as Apple's iTunes. I guess this problem is a great challenge for academics trying to find new branches in the field of Information Retrieval but the prospect of finding return on the investment is actually slim. After all, most applications that manage song libraries come built-in with powerful search engines that allow the average user find whatever she is looking for as long as she has a lead (label, composer, artist). iTunes goes one step further and allows the user to discover music based on the likes from other users. Something they call iMix.

For years we´ve been subjected to the influence of large marketing budgets pushing their favourite artists through the radio by paying handsome fees. The Top 10/100 lists are probably more influenced by those dollars than the actual preference of the average listener and yet that is most of what we hear on the street. Although in their infancy, technologies based on social networks such as iMix, will start creating a new way for the public to select their music by trusting the judgement not from the celebrity or radio DJ, but from the underground musician that just happens to have our same preferences.

But even trusting a complete stranger with the responsibility of feeding our brain with the newest and most eclectic music is way too risky. Tools will have to be developed that will allow simple metrics to be taken from a given song and produce a full profile of the chances that song is worth our time. This is where the technology I was referring at the beginning of this entry makes its appearance: by creating a digital signature for each creative segment within a song in such a way that they can be traced as they jump from song to song from time to time and contribute to make them popular or not. Although music is such a powerful language, these signatures could be accounted for and eventually we would be able to use this information to answer basic questions such as: will this song be of my liking based on my current library? or is this song a ripe from a previous hit, therefore doesn´t add to our culture? has this artist been influenced by another?

The application in question would probably look like a huge dependency tree where a few musical patterns seem to randomly appear all over the place but where truly creative sequences would have no match or precedence and could be clearly identified, just the same as simple imitations or samples would not do great. The process of decomposing a song in its most basic sequences and indexing them would not be too different to other applications where huge amount of data are scanned looking for particular repetitions.

This sort of application will become more critical as the popularity of flexible licensing options become more popular and allow for the sampling of someone else´s work to create new music. In the end, this is the only way to ensure that there are no ways of taking advantage.