Information Security for Automatic Speaker Identification

Abstract

Speaker identification is a widely used technique in several security systems. In remote access systems, speaker utterances are recoded and communicated through a communication channel to a receiver that performs the identification process. Speaker identification is based on characterizing each speaker with a set of features extracted from his or her utterance. Extracting the features from a clean speech signal guarantees the high success rate in the identification process. In real cases, a clean speech is not available for feature extraction due to channel degradations, background noise, or interfering audio signals. As a result, there is a need for speech enhancement, deconvolution, and separation algorithms to solve the problem of speaker identification in the presence of impairments. Another important issue, which deserves consideration, is how to enhance the security of a speaker identification system. This can be accomplished by watermark embedding in the clean speech signals at the transmitter. If this watermark is extracted correctly at the receiver, it can be used to ensure the correct speaker identification. Another means of security enhancement is the encryption of speech at the transmitter. Speech encryption prevents eavesdroppers from getting the speech signals that will be used for feature extraction to avoid any unauthorized access to the system by synthesis trials. Multilevels of security can be achieved by implementing both watermarking and encryption at the transmitter. The watermarking and encryption algorithms need to be robust to speech enhancement, and deconvolution algorithms to achieve the required degree of security and the highest possible speaker identification rates. This book provides for the first time a comprehensive literature review on how to improve the performance of speaker identification systems in noisy environments, by combining different feature extraction techniques with speech enhancement, deconvolution, separation, watermarking, and/or encryption.