Tamper Detection in Speech Based Access Control Systems Using Watermarking

Abstract

General voice based access control systems are based on voice biometrics. This process enables an unauthorized access by recording the voice of the authorized person. So there is a requirement to prevent unauthorized access through recording speech. Other than voice biometrics, here we have two challenges. (i) To extract the authentication information. (ii) To find the unauthorized source. The speech goes through DA-AD-DA conversion, while it is recorded and used for access control. The watermarking method which will use for this purpose must be robust to DA-AD conversion attack, which is usually involved in recordings. In this work, we propose a method based on casting Log Co-ordinate Mapping (LCM), in which embedding two watermark segments in two different frequency regions, one for authentication information purpose and other for finding unauthorized source. The LCM method has approving performance against DA-AD conversion attacks [1]. The modifications made for this does not impact the perceptible auditory quality and the embedding capacity improved by selecting the appropriate frequency regions in the log scale. Our results show that our method robustly extracts the source identification information while detecting the malicious source if the audio is being recorded and played back by unauthorized source.