Real Time Speaker Identification from Disguised Voice Using Feature Extraction and Classifications Techniques

Speaker identification is utilized in a multitude of applications, including but not limited to, voice-controlled software, systems for detecting telephone fraud, and information retrieval mechanisms. In this proposed research initiative, a text-dependent speaker recognition framework has been devised, which possesses the capability to classify the distinctive characteristics of speech signals and to recognize individual speakers. The feature extraction methodologies employed encompass Male Frequency Cepstral Coefficients (MFCC), Delta MFCC (ΔMFCC), and Double Delta MFCC (ΔMFCC), which are crucial for the extraction of features from both unaltered and manipulated vocal samples. Throughout the MFCC feature extraction procedure, the speech signal is segmented into frames, the Fast Fourier Transform (FFT) is computed, a mail filter bank is utilized, and subsequently, the Discrete Cosine Transform (DCT) is executed following logarithmic conversion. For the purpose of statistical analysis, the mean and correlation coefficients are derived employing MFCC and ΔMFCC feature extraction methodologies. Subsequent to the feature extraction phase, feature-based classifiers such as Support Vector Machines (SVMs) and Artificial Neural Networks (ANNs) are employed to evaluate classification performance. The outcomes of the classification reveal that the classification efficiencies achieved for the SVM and ANN classifiers are 93.33% and 95.77%, respectively. In contrast, previously established methodologies yielded classification efficiencies of 84.67% and 87.67% when utilizing SVM and ANN classification techniques. Upon conducting a comparative analysis, it becomes evident that the proposed model for speaker identification, which is predicated on feature extraction and classification techniques, surpasses existing methodologies in terms of performance.