Speaker Recognition Using Occurrence Pattern of Speech Signal

Speaker recognition is a highly studied area in the field of speech processing. Its application domains are many ranging from the forensic sciences to telephone banking and intelligent voice-driven applications such as answering machines. The area of study of this paper is a sub-field of speaker recognition called speaker identification. A new approach for tackling this problem with the use of one of the most powerful features of audio signals i.e. MFCC is proposed in this paper. Our work also makes use of the concept of co-occurrence matrices and derives statistical measures from it which are incorporated into the proposed feature vector. Finally, we apply a classifier which correctly identifies the person based on their speech sample. The work proposed here is perhaps one of the first to make use of such an arrangement, and results show that it is a highly promising strategy.