On the Selection of Audio descriptors and Identification of Singer in North Indian Classical Music

Abstract

Sound Information Retrieval has many subdivisions of research. Automatic Singer Identification is one of the most popular research sub-areas. Most of the research on singer identification is based on various ways in which the audio attributes can be extracted and the best possible ways in which the singer can be classified. The audio attributes are called as Audio Descriptors. An audio descriptor describes the information of an audio signal in compact and precise way. There are many types of audio descriptors. They are roughly classified into Temporal and Spectral features. Some of them are framing based (Instantaneous) and some are applied on entire audio signal (Global). There is no standard taxonomy available to neatly classify them into any category. Moreover, there is no standard rule to decide upon which and how many audio descriptors are to be used. Also, use of each audio descriptor differs from application to application. newlineAudio descriptors play important role in the applications such as Singer Identification, Musical Instrument Identification, Speaker Recognition, and Musical Genre Classification. Most of these applications are of type polyphonic or monophonic texture of sound. The process of identification becomes complex with respect to monophonic, polyphonic and homophonic type of the texture of the music. In North Indian Classical music (Homophonic version), an accompanying musical instrument called Tanpura, is continuously played during the vocal performance of the singer. In such cases, the standard procedures of singer identification are not sufficient. The contribution of audio descriptors changes from one type of audio input to the other. In this research we have proposed a novel method, Hybrid Selection Algorithm, to identify useful audio descriptors for Singer Identification in North Indian Classical Music. We have also proposed a weighing scheme to identify the effect of each audio descriptor on the process of Singer Identification.