TEXT-INDEPENDENT SPEAKER IDENTIFICATION SYSTEM USING PROBABLISTIC NEURAL NETWORK

Taif A. Mehdi; Mahir K. Mahmood

Taif A. Mehdi Electrical Engineering Department, College of Engineering, Al-Mustansiriya University.
Mahir K. Mahmood Electrical Engineering Department, College of Engineering, Al-Mustansiriya University.

Abstract

Text-independent closed set speaker identification system is achieved using a Probabilistic Neural Network (PNN) as a classifier and Reflection Coefficients(RC) as a speaker feature. The system is evaluated with a database consisting of 28 speakers(21 male and 7 female). Each speaker has three totally different sentences, the first is used for training and the rest are considered as a testing sentences. The system correctly identified all the database speakers when tested with noise free speech for the two test sentences. For 30 and 20 dB SNR noisy speech, the performance is almost unchanged. Keywords: Artificial Neural Network (ANN), Automatic Speaker Recognition (ASR), Cepstral Analysis, Probabilistic Neural Network (PNN), Speech Processing, Voice Biometric.