文章介绍的说话人识别系统，采用能够反映人对语音的感知特性的9(:频率倒谱系数（9(: <+(=>(2/?8(’0+): 81(.....

语音信号的共振峰抽取的算法-formant extraction in speech signal algorithm

配合《Matlab之语音处理与合成工具箱》的源代码库，书很经典，是英文原版。-With

利用HMM的语音识别代码，里面包含向前向后算法，重估baum算法，viterbi算法等，用C++编写的

marsyas是进行音频分析的源码，可以识别音乐节奏、乐器等音乐语义特征，使用隐马尔可夫模型。-marsyas is a source of audio analysis to identify the music rhythm, semantic features music and musical instruments, the use of Hidden Markov Model.

8kHz采样语音的基音检测程序，VC6.0环境下运行，附有matlab画图程序-8kHz sampling voice pitch detection procedures, VC6.0 environment running matlab with Paint program

这是一个通用的隐马尔科夫模型的C代码库， 非常好用。-This is a generic Hidden Markov Model C code libraries, very handy.

高质量的语音处理工具箱，包括分帧，预处理，求基频等-high-quality voice processing toolkit, including the frame, pretreatment, such as for frequency

数字音频格式有很多种，其质量与采样频率和采样精度两个参数有关。频率的表示单位为赫兹〔Hz〕，它表示每秒采样次数。采样频率越高，音质就越好。采样精度为每次采样所存储的数据数量，它决定每个数字信号所能够表示的离散振幅的数量。存储每个样本的数据越多，音质就越好。但是高品质的声音需要占用大量的内存和磁盘空间。考虑到网络带宽，在Internet连接上传输就需要花费很长的时间。对于Applet来说，保证声音文件的最小化是极为重要的。-digital audio format there are many, the quality and frequency of sampling and sampling accuracy of the two parameters. The frequency of Hertz unit of [Hz], it said the number of samples per second. The higher sampling frequency, the sound quality better. For each sampling precision sampling by the amount of data storage, it was decided that each digital signal can be expressed as the number of discrete amplitude. Each sample storage of more data, better sound quality. But high-quality voice would take a lot of memory and disk space. Consideration of the network bandwidth, the transmission on the Internet connection on the need to spend a very long time. Applet for instance, to ensure that voices of the smallest document is very important.

自适应量化。读取wav文件，使用2到12比特量化。抽样频率8khz-Adaptive quantization. Read wav files, using 2-12 bit quantify. Sampling frequency of 8kHz