Speech Watermarking Based on the Combination of Frequency QIM and Convolutional Coding

The frequent manipulation of speech information introduces an increasing copyright protection demand in the speech signal processing area. In this paper, we propose a blind speech watermarking method based on the combination of frequency Quantization Index Modulation (QIM0) and convolutional coding to enhance the robustness and watermarked speech quality. In the proposed approach, the input text message is transformed into a bitstream, then its convolutional encoded version is embedded into the frequency spectrum of the host speech signal according to the frame energy and the Human Auditory System (HAS). The detection process is performed on the frame that its energy is higher than a predefined threshold. The detected bitstream is then convolutionally decoded to obtain the watermarked message. The experimental results indicate that the proposed approach perform more robust against many signal processing attacks than the baseline methods. Furthermore, it can achieve more speech quality.