Speech Watermarking for Tampering Detection Using Singular Spectrum Analysis With Quantization Index Modulation and Psychoacoustic Model

Maliciously tampered speech can cause misinformation. A hidden watermark inserted into the speech, which will be destroyed if the speech is tampered with, can be used to detect tampering. Although several methods exist for encoding and decoding the watermark, the robustness, imperceptibility, and blindness still need to be improved. This study presents a method for detecting tampering in speech signals utilizing singular spectrum analysis with quantization index modulation (QIM) and a psychoacoustic model (PAM) to address the trade-off between robustness and imperceptibility and ensure blindness. The watermark is embedded by modifying singular values using QIM, which is recommended by PAM. The evaluation results show that the proposed method can extract watermark bits blindly, achieving results comparable to those of conventional methods. Additionally, sound quality increased by 14.80% in the objective different grade, 7.79% in the signal distortion ratio, and significantly reduced the log-spectral distance by 63.64%.