Have you ever listened to your favourite music and thought, ”I like those beats. Wish I could find songs with similar beats…!” Or maybe, “I wish I could just sort all my music according to the tunes...I would have made different collections for my every mood! Making playlist yourself is too mainstream! :P “
So my developer friends, this is very well possible! Yay!
Introduction
Let us discuss a relatively less explored field in the domain of Information Sciences which can help you develop such sorts of applications or models, Music Information Retrieval (MIR). As many of you less aware will think of it as new field emerging out of nowhere, but it has been there from about 2000s. Considering the history of music in our society, it is really surprising to see that MIR is so young. But it is growing since then at a fast pace. There are even proper courses to study MIR in the top universities around the world, like Stanford, MIT, to name a few.
Music information retrieval is actually a field of research with prime concern for extracting and inferring meaningful information from music, using these features for indexing purposes, and developing several retrievals and recommendation schemes. Now with increasing tread of MIR, several other tasks, such as fingerprinting, mood, beat tracking, cover song detection etc., are also being developed for.
Music is a multimodal human artefact. It can be represented in the form of audio, image, text, gesture and sometimes a mixture of these. According to Schedl et al. [ ], there several factors that affect the perception of music and its similarity, for humans, like beats, lyrics, friend’s suggestions or current mood of the person. Everyone perceives it in a different way. MIR techniques basically try to map the music to following categories: music content, music context, user context and user properties.
Categorization of perceptual music descriptors proposed in [3]
Aspects of music encoded in the audio signal, are referred as music content, whereas the elements which cannot be extracted from the audio music, like the culture of the artist, the cover image of the album, political scenario related to the music etc. belong to the category of music context. Looking at the user, user context again refers to the aspects of human life, such as emotions, surroundings and social call. However, user properties refer to the characteristics of the user like her taste in the music, or her education in it or perhaps her friend’s opinion about the singer.
Applications
As listed before, there are several core applications of MIR from user’s point of view.
Music Retrieval: As the name suggests, music retrieval applications help the user find music of her choice. This is done using some kind of similarity criterion. Fortunately, several kinds of research have been done for this purpose, providing us with lots of tasks and techniques to tinker with. The developer can perform tasks like fingerprinting, audio alignment, cover song detection, and a very interesting task of queuing by humming. The aim of all these tasks is the retrieval of music given certain criterion.
Music Recommendation: Like Amazon recommends a plethora of things to buy based on our shopping preference, similarly music recommendation systems presents a list of music items according to a user’s music preferences. A commercial example of this is last.fm, which is based on collaborative filtering model. Recent developments have come up with better systems which focus on multimodal recommendations.
Music Playlist Generation: Also referred as “Automatic DJing”, playlist generation somewhere relates to music recommendation. But an important difference lies in the fact that in normal recommender systems, the order in which user listens to the music doesn’t matter, but in playlist generation, it makes a huge difference.
Beyond these applications, there are several other uses of MIR that are being researched upon.
How do I begin?
The music information retrieval society has grown far and has several good books published by Springer and other renowned publishers. Moreover, the best thing one can do to step into this domain is to follow musicinformationretrieval.com, which contains a number of python tutorials on how to extract features of music, how to visualize data, analyse data, and many more.
Apart from this, many top conferences, like ISMIR (International Society of Music Information Retrieval), SIGIR etc. have research papers published in this field which can be very helpful in moving forward with MIR.
Conclusion
Beside many types of research and development in this field, there are many challenges yet to be resolved. There is still a need to better understand the user in MIR. Better descriptors need to form by digging deeper into the music itself. To build a complete system for MIR, researchers need to move beyond the horizon of the western music.
References
[1] musicinformationretrieval.com
[2] Markus Schedl, Emilia Gómez, Julián Urbano. Music Information Retrieval: Recent Developments and Applications, 2014.
[3] Markus Schedl, Arthur Flexer, and Julián Urbano. The Neglected User in Music Information Retrieval Research. International Journal of Journal of Intelligent Information Systems, 2013.
Comments
Post a Comment