Development of Speaker Verification System Using Three-Dimensional Feature

A speaker verification system purpose is to verify person's ID from their voice. Those voices are processed in the feature extraction process to obtain its voices characteristic or feature. Generally, the feature extraction process generates two-dimensional data as the voice's feature. In this paper, the features extraction is tried to output the voice's characteristic as the three-dimension data or as called as spectrogram. Because the speaker verification system needs a threshold value, in this paper, it is calculated using mean similarity score that resulted from the Euclidean distance value and structural similarity (SSIM) score. The Euclidean distance and SSIM are also used as the system classifier. The experiment shows that the verification applying the proposed method and Euclidean distance obtain 85.33% for the Euclidean classifier, and 82.67% for the SSIM classifier.