SimpleSpeech是一个使用HMM的自动语音识别系统。对象是研究的学生和对自动语音识别有兴趣的研究工作者。...

语音信号的频域处理,语音虽然是一个时变、非平稳的随机过程。但在短时间内可近似看作是平稳的。因此如果能从带噪语音的短时谱中估计出“纯净”语音的短时谱，即可达到语音增强的目的。由于噪声也是随机过程，因此这种估计只能建立在统计模型基础上。利用人耳感知对语音频谱分量的相位不敏感的特性，这类语音增强算法主要针对短时谱的幅度估计。 -voice signals in the frequency domain processing, voice is a time-varying, nonstationary random process. But in a short period of time can be approximated as smooth. So if Noisy Speech from the short-term spectrum estimate "pure" voice of the short-term spectrum, and reached speech enhancement purposes. As the noise is random process, which can only be estimated based on statistical models based on. Use ear perception of voice spectrum component of the phase sensitive to the characteristics of such speech enhancement algorithms targeted at the rate of short-term spectral estimation.

用VB写的呼叫中心-VB wrote Call Center

本源码为语音识别中lpcc（线性预测倒谱系数）的计算方法。经试验，可用。为方便大家理解，关键语句和函数后面都加上了注释。-The source code for speech recognition in lpcc (Linear Prediction Cepstrum Coefficient) is calculated. After testing, can be used. To facilitate the understanding of everyone, the key behind the statements and functions are added to the Notes.

SVM分类器，针对语音信号的分类识别应用。内容非常多，大家可以筛选着利用。-SVMclassification

cdma2000 1x语音编解码程序，并带有测试用例，很好用-cdma2000 1x voice codec procedures and with test cases, very good use

收集的ADPCM有关的资料，有源码，有原理说明，

语音识别方面的开发包，有助于设计HMM，NN和VQ。它是开放源码的。-the development kits, contribute to the design of HMM, NN and VQ. It is open source.

一个用vs2005开发的、简单的TTS文本转换语音程序，是.net版的TTS，有助于语音合成初学者。-One with the vs2005 developed a simple TTS Text speech translation procedure is. Net version of the TTS, speech synthesis will help beginners.

FSK demodulation telling how to do it after AWGN noise.

录音部分是参照vckbase的录音api代码,结合了一小段socket(TCP)就可以用来单方说话录音了,程序分两部分一部分是(录音机+网络发送代码),一部分是(接收数据+播放波形音乐代码),由于程序只是为了自己用,很多地方没有注意错误的识别,结构也比较乱,本程序当初最头疼的就是控制损耗内存,结果用了双缓存来存储波形数据来交替的存储/清除. -part of the recording light vckbase recording api code combination of a small socket (TCP) can be used for recording a unilateral words, part of the procedure is divided into two parts (recorders network code), part of the (receiving data waveform music broadcast code), as a result of procedures only for their own use. many places no attention to the wrong identification, the structure was chaotic, most of the first procedure is a headache for the control of memory loss, Results of the double-caching to store waveform data to the turn of the storage/removals.