Speaker Identification Using Convolutional Neural Networks and Mel-Frequency Cepstral Coefficients

The topic of voice-based person recognition is evolving rapidly, making advancement in deep learning to enhance biometric identification systems. This study presents a voice-based person recognition approach based on Convolutional Neural Networks (CNNs), with Mel-Frequency Cepstral Coefficients (MFCC) serving as the primary feature set. A unique dataset is created, consisting of voice recordings of 20 individuals, while 10 samples each. Our process begins with the extraction of MFCC features from the recordings, which is then used to train a CNN model for voice classification. The proposed model achieved an identification accuracy of 99.56 percent, demonstrating its effectiveness in distinguishing between different voices. The suggested model outperforms methods utilizing conventional machine learning techniques and performs competitively when compared to recent studies. The simplicity of proposed CNN architecture, combined with the robustness of MFCC features, provides an appropriate balance between computational efficiency and high accuracy.