To develop a robust speaker recognition system, the system should be able to provide acceptable performance with several operating conditions. A well-defined feature extraction algorithm makes the classification process more effective and efficient. This paper proposes a new method of identifying the speaker from speech signals using an artificial neural network. Here Mel frequency cepstral coefficient (MFCC) is investigated for feature extraction which gives useful features for the recognition process. Using this extracted features, input samples are created after its dimensions have been reduced using principal component analysis (PCA) for keeping the effective information and reducing the redundancy of characteristic parameters and also speeding up the training procedure. Datasets from FSDD of 3 speakers have been trained and tested using Multilayer Perceptron (MLP) trained with Genetic Algorithm (GA). Experimental results show that the method of using MLP-GA is more efficient to classify speech signals as compared to other methods.
Volume 11 | 04-Special Issue
Pages: 2451-2464