Investigation On Music Feature Representations For Data Retrieval

Noor Azilah , Muda and Norashikin , Ahmad and Sabrina , Ahmad and Noor Azilah , Draman (2012) Investigation On Music Feature Representations For Data Retrieval. Project Report. UTeM, Melaka, Malaysia. (Submitted)

PDF (Full Text) Investigation_On_Music_Feature_Representations_For_DAA_Retrieval_Noor_Azilah_Muda_QA76.87.I61_2013_DAH_EDIT.pdf - Submitted Version Restricted to Registered users only Download (8MB)

Abstract

Human capabilities for recognizing different types of musiC and grouping them into categories of genre are so remarkable that experts in music can perform such classification using their hearing senses and logical judgment. For decades now, the scientific community has been involved in research to automate the human process of recognizing genre of songs. These efforts would normally imitate the human method of recognizing the music by considering every essential component of the songs from artist voice, melody of the music through to the type of instruments used. As a result, various approaches or mechanisms have been introduced and developed to automate the process of classifying music genres. The results of these studies so far have been remarkable yet can still be improved. In order to do so, this research will investigate an Artificial Immune System (AIS) domain by focusing on the modified Negative Selection Algorithm to solve this problem where the highlight will be on a generalized shapespace concept called Hamming shape-space. In this highlight, threshold value will be emphasized where exhaustive search method will be applied to decide the threshold value that will provide the highest accuracy in music genre classification. Further, the identification method will involve four different similarity techniques to evaluate the Hamming shape-space concept in the modified negative selection algorithm.

Actions (login required)

View Item