Speaker recognition system based on pitch estimation

Automatic speaker recognition is to identify an individual from the audio recording of his voice. Several techniques exist in the current state of the art of the discipline. We designed a new technique comparable to those existing using the frequency of vibration of the vocal cords called the fundamental frequency. It is highly dependent on physiological characteristics of the individual. It is remarkably different from one person to another. We studied the existing techniques for estimating pitch and we chose the YAAPT technique (Yet Another Algorithm for Pitch Detection). Then we calculated the probability distribution of occurrence of each value of F0 in the speech signal to each speaker, and we modeled it by a Gaussian mixture (GMM). By testing our technique in text-independent mode and comparing it with other existing techniques in the literature, we noticed its performance.