Speaker Verification is a popularly used authentication mechanism for verifying the voiceprints of the identity claim for accessing the service. The speaker related speech features are the most important clues for recognizing a speaker in a speaker verification system. Protecting feature template is a challenging task against impersonation, eavesdropping attack and intra-user variability. Hence in the proposed system, feature extraction is done by using four feature extraction techniques and feature selection is done on the extracted features using optimization algorithm. The optimized features are watermarked by using bits of an image taken from the image database. The Least Significant Bit (LSB) of the selected image is taken and embedded with the considered features. Then, the watermarked feature set is encrypted by using the proposed Advanced Encryption Standard (AES). Finally, the proposed system is tested under various quality metrics to quantify the performance of the speaker verification system under statistical and brute-force attacks.

Login to Download PDF

Please login with your Paper ID and password to access the full PDF.

🔑 Login to Download