wvd_test

核磁共振（NMR）技术探测地下水是目前唯一的直接找水的地球物理方法。与传统的地球物理方法相比具有高分辨力，高效率，信息量丰富和解唯一性等优点。利用核磁共振地下水探测系统可以高效率地进行区域水文地质调查，确定找水远景区，圈定地下水在三维空间内的分布，进而可靠地选定水井位置等。但是由于水中氢核产生的核磁共振信号的幅度小（纳伏级），要求探测系统的灵敏度高，将引入大量的自然和人为噪声，导致采集信号的信噪比低，解释结果不清楚。 本文用LMS自适应算法提高MRS信号的信噪比，用Hilbert变换提取MRS信号的包络信号，用线性拟合提取水文地质参数并通过仿真实验结果，验证了本文提出的数据处理方法的有效性和准确性。 (Nuclear magnetic resonance (NMR) technique to the detection of direct water division is the only geophysical method to find water. Compared with conventional high-resolution geophysical methods, efficient, informative unique advantages of reconciliation. Groundwater exploration using nuclear magnetic resonance system can efficiently carry out the regional hydrogeological investigation to determine the prospective areas to find water, delineation of ground water distribution in three-dimensional space, and then reliably the location of the selected wells. However, because the water proton NMR signal amplitude generated by small (Na V level), requiring high sensitivity detection systems, will introduce a large number of natural and man-made noise, resulting in low signal to noise ratio acquisition, interpretation of results is not clear. In this paper, LMS adaptive algorithm improves the MRS signal to noise ratio, with the Hilbert transform to extract the signal envelope MRS signal ext)

利用GMM训练模型，并用于说话人识别等相关领域，有不错的效果(GMM apply for text to speech)

说话人性别识别matlab程序，基音检测、共振峰检测、GMM模型识别(Speaker sex recognition matlab program, pitch detection, formant detection, GMM model identification)

LMS多麦克风语音降噪的主程序是lmsspdn.m(Multi-microphone noise reduction LMS voice is the main program lmsspdn.m)

这里面的代码是2DPCA的。这篇文章发表在PAMI上,Two-Dimensional PCA: A New Approach to Appearance-Based Face Representation and Recognition.google搜索已经引用681次。这里的代码是原作者的。这个分类不太好，语音识别是最近的，这是模式识别里的经典算法。(matlab code of 2DPCA)

用MATLAB来实现谱减法降噪。有源程序和降噪先后的语音信号，有实验的结果图(Use MATLAB to achieve spectral subtraction noise reduction. There has voice source and noise signals, there are experimental results of Figure)

gmm模型下的语音情感识别系统，GMM只是一个数学模型，只是对数据形态的拟和，但是和你所看到的数据分布存在出入也是正常的，因为用EM估计GMM的那些参数时，一般假设我们所得到的数据是不完备的（也就是说假设我们看到的数据分布不是真正的分布，它在运算时把那部分丢失或者叫隐藏的数据“补”上了）(gmm model speech emotion recognition system, GMM is a mathematical model, but fitting the data form, but you can see there is access to the data distribution is normal, because those with the EM estimated GMM parameters, the general assumption we get the data is incomplete (that is, assuming we see the distribution of the data distribution is not real, it is part of the operation at the time that is lost or hidden data is called " fill" on the))

HMM隐马尔科夫模型源码，实现模型的训练(Hidden Markov Model HMM source, the realization of model training)

attention model for saliency detection.

观测实时模拟信号（语音）的频谱 用音频设备采集一段语音，将语音存为.wav格式。对wav文件作分段傅里叶变换分析。语音是分音节的，应把它分段分析，而且实际运用中的数字信号处理的FFT的点数是有限的，一般只能达到千点。用傅里叶反变换IFFT，从频域恢复信号。画出频谱图和语音波形图。 (Observing real-time analog signal (voice) of the spectrum collected with the audio devices section of voice, voice saved as. Wav format. Wav file of the sub-Fourier transform analysis to make. Speech is divided into syllables, it should be sub-analysis, and practical application of digital signal processing of the FFT points is limited, generally only reach 1000 points. Fourier inverse transform IFFT, to restore the signal from the frequency domain. Draw the spectrum map and voice waveform.)