SPEAKER RECOGNITION SYSTEM WITH MFCC FEATURE EXTRACTION AND NEURAL NETWORK BACKPROPAGATION

Abstract

Sound is the identity of human beings who are unique and inherent in the human body. Human speech recognition system using human voice that was extracted by the MFCC method, generating matrix and stored in a database. The process of identifying the human voice with voice to match tested and matched to the matrix that exists in the database using artificial neural network algorithm. A number of sound 30 files of 300 people. A sound file in a matrix waveread 2 x 176520 then extracted with MFCC be 10 x 501 (10 Chanal) further compression by PCA method to be 10 x 4 per voice and direshape to 1 x 40 (change column so rows) the end result of the extraction of sound before entering the neural network 40 x 300. Sound pattern recognition using artificial neural networks have a 80% accuracy rate.