Week 5 Blog 3: Automatic Analyzing Songs Using Machine Learning Algorithms

Automatic analyzing songs
Computers have made big changes in musical industry. Right now, people can find music by their title, type of genre, name of artist, but they are not able to recommend a good tune. There are plenty of recommendation systems for music listener. For example iTunes recommends songs to the user based on comparing the collection of other user. Now researchers are using machine learning algorithms, “in combination with a Facebook game, to classify music based on the automated analysis of the songs.” The researchers want their system to be able to not only tag songs with title or artist but they will be able to tag by genre, instrument, and descriptive words. Researchers are gathering information by using a game called Herd It. I believe in order to encourage people to play this game, they will award users points when they tag songs in a similar way to other user’s tagging. Using most of machine learning algorithms, in order to give the satisfy results, required lots of sample data. Once the data is collected, first they group the songs, it’s given the same tag by users and then applies a statistical analysis to the waveform patterns, and they will look for common feature between the songs with same tag. The researchers believe that the system has commercial potential. The system expected to take an unfamiliar song, from anywhere, and suggest appropriate tag and similar music. Lots of companies who were even trying to reach to this goal long time ago, before this research like Pandora, the Internet radio site, has done this but 100% manually by getting help of their users. Other company like the Echo Nest is working on automatic analysis of music. They are transforming the waveform patterns of songs based on what is familiar to human ear and then combining that information with tagging information gleaned from blogs and other posted on the Internet. Then they apply machine learning algorithm to identify feature of songs that are associated with specific tags. As researchers expect the automatic analyzing the song will increase as the demand for song is going up and sites will encourage users to tag and find new music in order for the researchers get the better results in their field of research by using machine learning algorithms. I as one of the user who do not listen to music that often or not looking for any special or new music that are coming out because of not being familiar with singer’s name or the albums. I believe this research and the system that are coming up with will encourage me to look for music and listen to it. Because I believe this system even help the user to save the time and make decision much faster. The usage of this algorithms will make searching much faster and gives more accurate recommendation.

Source:
http://www.technologyreview.com/computing/22894/page2/

Other Sources:
http://webhome.cs.uvic.ca/~gtzan/mir_course/papers/FXPAL-PR-02-171.pdf