Automatic Speaker Recognition Using Mel-Frequency Cepstral Coefficients Through Machine Learning

Automatic speaker recognition (ASR) systems are the field of Human-machine interaction and scientists have been using feature extraction matching methods to analyze synthesize these signals. One most commonly used for is Mel Frequency Cepstral Coefficients (MFCCs). Recent researches show that MFCCs successful in processing voice signal with high accuracies. represents a sequence signal-specific features. This experimental analysis proposed distinguish Turkish speakers by extracting from speech recordings. Since human perception sound not linear, after filterbank step MFCC method, we converted obtained log filterbanks into decibel (dB) features-based spectrograms without applying Discrete Cosine Transform (DCT). A new dataset was created spectrogram 2-D array. Several learning algorithms were implemented 10-fold cross-validation method detect speaker. The highest accuracy 90.2% achieved Multi-layer Perceptron (MLP) tanh activation function. important output this study inclusion as set.