Robustness Improvement against G.726 Speech Codec for Semi-fragile Watermarking in Speech Signals with Singular Spectrum Analysis and Quantization Index Modulation

Semi-fragile watermarking in speech signals is proposed to solve problems relating to unauthorized speech modification. However, previous methods are fragile against some non-malicious attacks or white noise with a high signal-to-noise ratio. This paper aims to solve this problem by proposing a new watermarking technique based on singular spectrum analysis and quantization index modulation. The singular spectrum analysis is used to extract singular values of segments of speech signals. A watermark bit is embedded into each frame by slightly modifying its singular values according to the quantization index modulation. The experimental results show that the sound quality of a watermarked signal is comparable to that of its original signal. The watermark-bit extraction precision is also similar to that of existing methods. However, the proposed method is robust against non-malicious attacks, such as G.726 speech codec and white noise with a high signal-to-noise ratio.