Digital Watermarking Based on Process of Speech Production

摘要

A speech production model can be divided into three parts, namely the glottal source, articulation and radiation, respectively. Some digital watermarking methods for speech that have been proposed are based on modifying quantized values or parameters of a coding scheme. In this paper, we propose a new watermarking method for speech by manipulating the articulation in the process of speech production. The proposed method is performed by modeling a quasi vocal tract model equivalent to the speech production process. The watermarked vocal tract model is expressed by codebooks made by LSP (Line Spectrum Pair) parameters. The procedure of watermark for speech is as follows; 1) LSPs are extracted from the speech. 2) Some of the extracted LSPs are replaced by the codebook vectors. 3) Speech is synthesized using replaced LSPs. In the process above, watermarks are embedded indirectly into the speech. Evaluation tests on speech quality and accuracy of the proposed method will be discussed with simulation results.