A two-level approach for speaker recognition using speaker-specific-text

Abstract

In speaker recognition tasks, one of the reasons for reduced accuracy is due to closely resembling speakers in the acoustic space. In conventional GMM-based modeling technique, since the model parameters of a class are estimated without considering other classes in the system, features that are common across various classes may also be captured, along with unique features. If the system is designed to use only the unique features of a given speaker with respect to his/her acoustically resembling speaker, then the system is expected to perform better. In this proposed work, the effect of a subset of phonemes, which are unique to a speaker, in the acoustic sense, on a speaker identification task is investigated. This paper proposes a two-level approach to reduce the confusion errors, by finding speaker-specific phonemes and formulate a text using the subset of phonemes that are unique, for speaker identification task. Experiments have been conducted on speaker identification task using speech data of 50 speakers collected in a laboratory environment. The experiments show an increased accuracy for the proposed two-level classifier when compared with that of a conventional GMM-based technique.