Matlab环境下的基于独立分量分析（ICA）算法的声音混合与分离仿真程序-Matlab environment based on the independent component analysis (ICA) is the voice mixed with the separation of simulation program 下载
|
语音识别中的模型和算法：动态时间归正技术（DTW），隐马尔可夫模型（HMM），高斯混合模型（GMM），高斯混合模型（GMM）-Speech Recognition model and algorithms : dynamic time Reformed technology (DTW) Hidden Markov Model (HMM), Gaussian mixture model (GMM), Gaussian mixture model (GMM) 下载
|
自己编写的一款可以进行双路PCM语音混频的程序，针对的是16位的PCM语音数据，经过验证，比常规的混音算法质量好许多，特别是进行多路混频时，重复调用该程序也可以达到比较满意的结果。-own section can be prepared for two-way voice PCM mixing procedures, The address is 16 PCM voice data, after verification, than conventional mixing algorithm many good quality, especially for multi-channel mixer, repeat the procedure call also will be more satisfied with the results. 下载
|
在55XDSP下写的二阶累积量盲分离程序，算法运算量小，包括片内PLL初始化，并口初始化。可实时分离瞬时混合语音信号。-55XDSP in writing under the second-order cumulant blind separation procedure, a small amount of computation algorithm, PLL including chip initialization, Parallel Initialization. Isolation can be instantaneous real-time voice mixed signals. 下载
|
在linux平台下,纯c写的盲信号分离的代码.它采用基于卷积混合的盲信号分离算法,不但可以分离人工合成的混合信号,而且对于真实环境中的卷积混合的语音信号也能够分离.在本程序中,包含了两个测试文件,makefile后便可以直接使用. 另外值得一提的是,压缩包里包含有一些语音处理方面的常用函数.例如fft变换,读取\写入wav文件,以及一些常用的一维向量和二维矩阵变换的函数.这些可以直接应用在其他应用程序里去.-in linux platform, net write c Blind Signal Separation code. it is based on convolution of mixed Blind Signal Separation algorithm, not only can be separated from the artificial synthesis of mixed-signal, but for the real environment convolution of mixed voice signals can be separated. in this procedure, the tests included two documents, makefile can be used directly after. Also worth mentioning is that Compression bag containing some of voice processing functions commonly used. fft transform, for example, to read \ write wav file, and some of the commonly used one-dimensional and two-dimensional vector matrix transformation functions. these can be directly applied to other applications going. 下载
|
网页是组成互联网的基本数据单元，是各种面向互联网的应用系统最原始的数据源。网页内部含有大量噪音信息，如何从网页中有效地提取有价值的内容成为影响数据处理效果的关键。 网页正文提取指的是从原始网页中精确地提取出正文文本，比如提取新闻网页中的报道内容。能否高效地提取出网页的正文，是很多互联网应用系统如搜索引擎、新闻资讯系统等面临的一个重要问题。由于网页本身的无结构化的特点，通常采用的正文提取方法是针对目标网页的特点人工制定抽取模板，这类方法的优点是抽取精确，但其致命的缺点是模板建立和维护的工作量巨大，通用性和灵活性很差。 通过分析网页内部的链接分布特点，我们研制出了一种基于网页上下文链接密度的混合型正文判定算法，能够有效地解决上述通用提取方法的缺点，其最大特点是无须模板支持，因此不需要人工制定抽取和维护模板，将人工参与的工作量降到0。另外该方法具有很好的提取效果，对新闻网页的测试表明，该方法的准确率和召回率都在98％以上。 -Internet website is composed of the basic data units, is the Internet-oriented application system the most primitive data sources. Internal website contains a lot of noise information, from the website how to effectively extract the valuable contents of the data-processing become the key. Website text refers to the extraction from the original website accurately extracted the body text, such as news from the website as reporting. Can efficient extraction of the body of the website is that many Internet applications such as search engines, News and other information systems facing an important issue. As the website itself without structural characteristics, commonly used text extraction method is targeted website features developed from artificial template The advantages of these methods is 下载
|
本文完成了对唇动身份识别技术几个基本问题的理论研究,并对整个系统加以实现.作为本文研究的实验基础,我们建立了唇动方式身份识别数据库(HITLUDB), 该库目前包含30个说话人每人20个汉语词的音视频语料.数据库的扩充与完善工作仍在不断的进行之中.在嘴唇检测方面, 我们对自适应色度过滤模型进行改进,提高了算法的鲁棒性,完成了对嘴唇的精确定位.结合DCT变换与K-L变换的各自特点, 我们提出了特征提取算法,使用较少维数的特征完成了对嘴唇区域主要信息的刻画.由于唇动信息同时包含了生理特征与行为特征, 我们使用静念动念混合建模的方式,完成了对说话人唇动个性特点的精确描述.在HMM训练时,我们提出了特征的归一化处理方法,提高了HMM在实际应用中的性能. 最后,我们分别对身份辨认系统与身份确认系统的基本理论进行了叙述,并完成了系统的实践工作. 关 键 词：身份识别 唇动 特征提取 隐马尔可夫模型 K-L变换 -paper completed the lip movements identification technology several basic issues of theoretical study, system as a whole be achieved. As this paper, the experimental basis, We have established a dynamic manner lip identification database (HITLUDB) The library currently contains 30 words each of 20 Chinese words, sound and video corpus. and the expansion of the database is still perfect keep going on. Detection of the lips, we adaptive color filter model improvements, improve the robustness of the algorithm, completed a pair of lips the precise positioning. DCT combined with the K-L transform their own characteristics, We have proposed a feature extraction algorithm, use less dimension of the lips completed the main message of regional characterization. As the lip movements of information, 下载
|