【作者】 时磊；

【导师】 尚秋峰；

【作者基本信息】 华北电力大学（河北） ， 通信与信息系统， 2008， 硕士

【摘要】 本文在分析了语音信号声学特性、感知特性的基础上,针对小波和神经网络在语音处理中的优良性质,提出了使用小波和BP网络相结合的方法压缩语音数据。文中将整个语音压缩系统分为两个部分:小波模块部分和神经网络模块部分。在小波模块中,语音首先经过小波变换,转化为小波系数,然后通过小波阈值处理,压缩语音数据中不重要的信息,之后对包含重要信息的小波系数进行量化编码;神经网络模块中,利用小波模块产生的二进制数据作为神经网络的目标输出,同时产生出固定的矩阵数组作为输入。网络训练完成后传递网络的权值和阈值,达到神经网络压缩的目的。仿真实验表明,在保证语音质量可听的情况下,压缩倍数可以达到31倍左右。更多还原

【Abstract】 This article analyses speech signal acoustics characteristic, perception characteristic. Considering good character of the wavelet and neural networks in speech treatment, the method of integrating wavelet and BP neural networks is submitted to compress speech data. Speech compression system is divided into two parts: wavelet module and neural networks module. In the wavelet module part, speech signal is changed into wavelet coefficient at the first time. Then unimportant information of wavelet coefficient is compressed through wavelet threshold value. At last the important information of wavelet coefficient is quantized and encoded. In neural networks module part, the output is the binary data from wavelet module. At the same time fixed matrix is generated to be the input of neural networks. When the network is trained to be completed, the threshold value of the network is preserved. At last the threshold value is transferred and compression data purpose is reached. The simulated experiment is indicated that speech signal can be recognized when the compression ratio is around 31.更多还原