Speaker Identification System Using Artificial Neural Network By Enhancing the Signals
Khine Zin Oo, Lwin Nyein Thu · 2024
Speaker identification (SI) is the progress of classifying a person according to their voice and is now a supported topic for forensic research. However, the performance of SI research is related to the quality and size of the speech signal being performed. The primary aim is to increase the effectiveness and precision of a proposed system when background noises are present. In order to perform the analysis, the first step to get significant features is to enhance the input audio files. To produce enhanced signals, active voice detection (AVD) is thus used before characteristics are extracted. In this paper, several AVD methods are applied to get the only useful parts of the input signal, both Mel-Frequency Cepstral Coefficients (MFCC) and Discrete Wavelet Transform (DWT) of feature extraction methods are conducted to get proper features, and artificial neural network (ANN) is implemented to identify the speaker. On our created dataset of Myanmar Speakers, experiments are carried out. The study results compare the accuracy of models using both the original signal and enhanced signal with both feature extraction techniques.