With the incredible development of network technology, audio piracy and malice fiddle become increasingly rampant, so the digital audio watermarking technique has been payed more attention day by day. The digital audio watermarking technique is a crossover subject which including communication theoryã€cryptologyã€chaos and digital audio. It shows extensive application outlook in computer, communication and secrecy and many other fields. Recently, the researches about digital audio watermarking have rapidly improvement in recent years.This paper makes systematic researches and discussions about audio watermarking in the following facets:the theoretic basisã€algorithm constructionã€performance optimization and watermark detection. Some important researches have been made aiming at zero watermark algorithm of embedding meaningful watermark, attack sensibility of fragile watermark and modifying accurate positioning of semi-fragile zero watermark. The main creative results of the paper are described as follows:1. Algorithm model of audio zero watermark based on neural network is built up. The model utilizes neural network to embed meaningful zero watermark into raw audio. Not only the contradiction between imperceptibility and robustness has been effectively solved, but also more intuitionistic meaningful watermark is embedded. So, the problem that current zero watermark is only capable of being embedded in binary watermark is solved. During extraction and detection of watermark, only key is needed to extract feature data and watermark blind detection is realized by machine learning. BP neural network is applied in this model of audio zero watermark. Attack tests shows well concealment of the zero watermark schemeã€simple watermark detection method and good robustness against diversified attacks.2. Ant algorithm is adopted to optimize the original weight values of BP neural network, which is used in the audio zero watermarking model Ant algorithm based on ant heap formation can be used for data clustering. But it has the drawbacks of slow convergence rate and probability searching. So the algorithm is optimized in this paper. This paper put forward new similarity function, pick-up and drop-down probability function. It fastens the convergence rate of algorithm. The optimized algorithm can also make the results posses the ergodic property. Optimized ant algorithm is used for weight values optimization of BP neural network and model of audio zero watermark based on neural network. Experimental results further demonstrate effectiveness of optimized algorithm. The concrete embodiments are as follows:decreases in average error of BP neural network, remarkable decreases in training time, further increase in zero watermark’s robustness against attacks.3. An audio dual-watermark algorithm based on bipolar quantization is proposed. The robust watermark is constructed by zero watermark of above ant optimized BP neural network, meanwhile, the fragile watermark adopts the new bipolar quantization method presented in this paper. The advantages of new bipolar quantization method lies in the following aspects:difference of parity coefficients of wavelet packet decomposition is selected as quantization object, take interval’s boundary values instead of middle values which is used in regular quantization, the watermark is scattered to odd coefficients of wavelet packet domain, the sensibility of fragile watermark is improved, meanwhile, the distortion type can be identified, and moreover, accurate positioning can be made to malicious distortions.4. A semi-fragile audio zero watermark algorithm is presented. The algorithm extracts the raw audio signal’s middle and low frequency components as features and constructs semi-fragile zero watermark by connecting computation with watermark image. Experimental results show that:this semi-fragile audio zero watermark have small amount of computation and is easily realizedã€capability of differentiate regular attack and malicious fiddle〠anti-attack robustness against regular signal processing which maintain content operation. When making analysis about malicious fiddle, the location positioning is done to malicious fiddle firstly, then the accurate positioning is processed, morever the destroyed field of raw audio content is deduced, finally the attack purpose of malicious fiddler is confirmed. |