9.1 Abstract and goal of the genre detection method

Abstract and goal of our genre detection method

Abstract of our genre detection method

Categorizing music isimportant to all music fans nowadays. However, few programs currently existing on themarket can automatically, efficiently, and accurately detect the genre of a random chosen song. For example, before customers cansearch for a list of songs within a certain genre on iTunes, the built-in genre detector for the widely-used ipod series, they haveto manually input each song’s genre in order to make it happen. Thus by increasing the accuracy and shortening the detection time,we can come up with a really useful and cool project that will ensure higher quality and customer satisfaction.

Goals

The goal of our genre detection project is to develop a algorithm that detects the genre of any song. Thespecific genres (our testing pool for now) that we are seeking to detect are rap, classical, rock, jazz, and pop.

Methods of detection

In our project, we have developed two different techniques to complete the detection process. For both,we created a database “comparison” matrix composed of certain information of the previously mentioned genres. Then in the sameway we composed another matrix containing information of the input song and compared it with the database to find the most similargenre. The first technique was with the fast Fourier transform and matrix multiplication. We used matrix multiplication and the dotproduct to find the similarities between our input song and each genre. The second technique involved finding linear predictivecoefficients (LPC), which finds the predicted nth value. The LPC detects genre was based on error while the FFT matrix methodachieved the same goal by finding the similarity.