Feature selection method for music mood score detection

Genre classification of songs using neural network

“…Comparing the results with the previous work, the accuracy acquired is quite high as compared to the work done in [1], [5] , [7] and [6] but nearly same as in [4]. But as compared to [4] the feature set has been greatly reduced from 174 vectors to just 34 which clearly indicate that lesser amount of time, cost, effort and energy would be used to build this system.…”

“…In [7] mood of the song plays a good role in the classification of songs. To make our model more accurate valence i.e., mood of the song is also derived.…”

“…It determines the positivity displayed by the song i.e., cheerful, happiness, sad, depressed or angry. While calculating these factors as stated in [7] certain assumptions are made. The assumptions that a high energy frame represent a cheerful or brightness while a low energy frame is supposed to be a quite frame.…”

“…The assumptions that a high energy frame represent a cheerful or brightness while a low energy frame is supposed to be a quite frame. With these assumptions and taking the log spectrum sum of the power spectrum the value determined as in [7] is used to determine the valence of a song.…”

“…This may be because MFCC are mainly used for speech recognition. MFCC can be said as an artificial human ear as it was designed in consideration of the human auditory system.The work by Masato Miyoshi is based upon the mood of the song [7] i.e., valence was proposed which is an interesting feature that can be extracted from a song but what about a sad Rock and Pop song. The classifier based on the mood will surely put both these two different genre songs under the same title which will not be acceptable to a genre classifier as mood is only detecting what type of song i.e., happy, sad, cheerful, etc., it belongs to.…”