Tampering Detection for Speech Signals Using Synchronization Code and LSF-based Watermarks

Speech tampering has become a serious threat to speech signals. This paper proposes a method to detect horizontal and vertical speech tampering. In this method, each speech frame is divided into two non-overlapping parts to embed the synchronization code and watermarks, respectively, where the synchronization code is used to identify horizontal tampering and the watermarks are used to detect vertical tampering. To increase the robustness and security of the frame synchronization, the synchronization code (a random binary code) is embedded into the wavelet domain. Efficient line spectral frequency (LSF) information is extracted from the synchronization part and then embedded into the watermarks part using echo-hiding. Tampering detection is achieved by comparing the embedded LSF information with the extracted LSF information. Experimental results demonstrate that the proposed method provides satisfactory inaudibility, robustness, the fragility, as well as high accuracy for tampering detection with a detection precision of 0.1 s.