Security against morphed speech based on speaker verification system

摘要

Speaker Verification systems needs to be robust over Morphed or Converted voice. There is possibility of Spoofing attack so to prevent that kind of attack and make surety of the system is big challenge. The main goal of the proposed system is to correctly identify the speaker and preventing the system from the morphing attack to ensure the authenticity, by using security. Morphing is performed by Pitch Synchronous Overlap and Add (PSOLA) method and Verification is performed by Correlation method, in that feature extraction is performed by MFCC and LPC method. Watermarking is performed with the blind detection and DM-QIM is used for watermarking process. In order to evaluate the overall performance of implemented algorithm, Identification rate, ABX test, and WSR are used.