基于LP-MMSE的高效语音感知哈希认证算法

An efficient speech perceptual hashing authentication algorithm based on the linear prediction minimum mean squared error

为了满足移动语音通信中对语音内容保持操作的鲁棒性和区分性,并实现高效认证等要求,提出一种基于线性预测最小均方差(LP-MMSE)的高效语音感知哈希认证算法.该算法首先对待认证语音进行预处理、分帧和加窗后的语音信号进行线性预测分析(LPC)得到其最小均方差(MMSE)系数矩阵;然后对分帧后的语音利用谱熵法计算得到每帧的谱熵值参数矩阵;最后结合上述两个矩阵,生成最终的二进制感知哈希序列.仿真结果表明,该算法的感知哈希摘要性优于现有的其它方法并对内容保持操作具有较好的鲁棒性和区分性,认证效率高,能够满足语音通信实时认证的要求.

In order to meet robustness and discrimination of content preserving operations in mobile speech communication ,and the need for high-efficiency certification and other requirements ,a novel efficient speech perceptual hashing authentication algorithm based on the linear prediction-minimum mean squared error （LP-MMSE） was proposed .Firstly ,the algorithm conducted linear prediction coding （LPC） on speech signal after pre-processing ,framing and windowing to obtain the minimum mean squared error （MMSE） coefficient matrix .Secondly ,spectral entropy parameter matrix of each frame was calculated through the spectral entropy method .Finally ,binary perceptual hash sequence was generated through combining these two matrixes above .Experimental results show that the pro-posed algorithm is better than other existing algorithms in terms of compactness ,and has a good ro-bustness ,discrimination against content preserving operations .It also has high authentication effi-ciency and can meet the requirements of real-time speech authentication .