Skip to Main Content
A fragile speech watermarking scheme of low complexity is proposed in this research as an effective way to detect malicious content alterations while tolerating content preserving operations. Since no auxiliary data are required, the authentication process is transparent to agents that perform content preserving operations such as transcoding during the transmission process. The proposed scheme is based on the modified odd/even modulation scheme with exponential scale quantization and a localized frequency masking model while assuring no mismatch between quantization steps used in watermark embedding and detection. The noise introduced by watermarking is shown to be smaller than that of speech coders. It is shown by experimental results that the proposed system is able to distinguish malicious alterations from resampling, white noise pollution, G.711 speech coding and G.721 speech coding with very low error probabilities.