搜索资源列表
getparam
- 在基于hmm的语音识别中计算前向概率,后向概率以及各种参数结构的源代码。-based on voice recognition to the calculation of the probability that after the probability of various parameters of the structure of the source code.
Wordsegmentation2
- NLP技术实现,对语料库进行自动统计生成分词词典,对训练集进行分词,列出所有的分词可能并计算每种可能的概率。请使用者自行加入语料库和测试集。-NLP technology to automatically Corpus Health Statistics ingredients dictionary, the training set for segmentation, list all the sub-term may calculate the probability of each pos
mypdfgauss
- 一个关于高斯模型的概率分布图的生成的小程序,欢迎下载-a Gaussian model of the probability distribution graph generation of small procedures, welcomed Download
forward_viterbi
- The algorm of viterbi. You talk to your friend three days in a row and discover that on the first day he went for a walk, on the second day he went shopping, and on the third day he cleaned his apartment. You have two questions: What is the overall p
20090226
- 从盲声源信号的独立性出发!提出了一种新的盲声源混合信号分离方法:该方法基于信号联合概率的 分布统计!利用信号联合概率的方向导数熵最小获得最佳的旋转角度!最终实现盲信号分离:与快速独立分 量分析方法及神经网络方法相比!该方法不需要迭代计算:采用新的盲声源信号分离方法对轴承试验台的混 合声音信号进行识别!将电机和滚动轴承的声音分离出来!进而可以准确识别机械的故障-Blind sound source from the independence of the starting signal
yuyinxinhaolianghua
- 采用矢量量化对语音信号量化,并与标量量化对比分析,绘制了语音信号的概率密度曲线-The use of vector quantization for speech signal quantization and scalar quantization and comparative analysis, rendering the speech signal probability density curve
HMM
- :为了使应力变异在顽健语音识别系统中能够达到较好的识别效果,研究了基于隐马 尔可夫模型(HMM)的自适应技术,提出了将最大后验概率(MAP)和最大似然回归方法(MLLR)用 于应力变异语音的自适应中。实验结果表明,与基本系统相比,两种方法均有效地提高系统识别 率。以SD为初始模型的最大后验概率方法在150个训练样本时识别效果最好,可以达到90.4% 。-: In order to stress variation in the robustness of speech recogni
132413241234
- What is the probability that the weather for the hmm
SpeechProbabilityDistribution
- Speech Probability Distribution.
speech2
- 为了提高语音分离算法的收敛速度以及分离性能,提出把拉普拉斯正态混合分布概率密度函数作为语音信号概率密度函数的估计,得到一个更加适合语音信号分离的激活函数,基于此函数提出一种快速语音分离算法.-In order to improve speech separation algorithm convergence speed and separation performance, raise the normal mixture distribution Laplace probability de
Unsupervised_Adapting_in_Speech_Recognising_using_
- 介绍了一种基于词网的最大似然线性回归无监督自适应算法,并进行了改进。根据解码得到的词网估计变换参数,词网的潜在误识率远小于识别结果,因此可以使参数估计更为准确。传统的一个很大缺点是计算量极大,较难实用,对此本文提出了两个改进技术:1利用后验概率压缩词网;2利用单词的时间信息限制状态统计量的计算范围。实验测定,误识率比传统相对下降了。-Introduced the term network based maximum likelihood linear regression unsupervise
Spedaker_Adapting_in_Speech_recognizing
- :自适应技术在近年来得到越来越多的重视,其中应用广泛的包括,-.、,//0,该技术利用少量特定 人数据就可以调整码本,快速地提升识别性能,它要求原始的码本有很好的说话人无关性。本文介绍了结合 ,//0 自适应的说话人自适应训练(1234536 -74289:3 649<9<=,以下简称1- )算法,这种方法将每个说话人码本 视为说话人无关码本经过线性变换的结果,在此基础上训练的说话人无关码本更有效剔除了说话人相关信 息,因此在说话人自适应中时能根据特定数据调整更好地逼
GAUSSIANloglikelihood
- GMM高斯混合模型大规模概率对数计算 需要一个模型地址文件和一个需要识别的声音的mfc文件可以一次执行大批量-GMM Gaussian mixture model probability on the number of large-scale computing need a model of address file, and the voice of the mfc file which need to be identified .can be an implementation
hmm
- hmm文件时运用HMM算法实现噪声环境下语音识别的。其中vad.m是端点检测程序;mfcc.m是计算MFCC参数的程序;pdf.m函数是计算给定观察向量对该高斯概率密度函数的输出概率;mixture.m是计算观察向量对于某个HMM状态的输出概率,也就是观察向量对该状态的若干高斯混合元的输出概率的线性组合;getparam.m函数是计算前向概率、后向概率、标定系数等参数;viterbi.m是实现Viterbi算法;baum.m是实现Baum-Welch算法;inithmm.m是初始化参数;trai
Bayeslilun
- 贝叶斯决策理论方法是统计模式识别的一个基本方法,类概率密度已知和类别数已知-Bayesian decision theory method of statistical pattern recognition is a basic method, the class probability density known and category known number
voice-box-GMM
- 语音处理GMM相关算法,1.计算概率密度并画出高斯混合模型,2.计算边际,条件混合高斯密度,3估计两个GMM模型的Kullback-Leibler divergence。-GMM relating to speech processing algorithms.1,to calculate probability densities from or plot a Gaussian mixture model.2,marginal and conditional Gaussian mixture
Speech Encoding - Frequency Analysis MATLAB
- The speech signal for the particular isolated word can be viewed as the one generated using the sequential generating probabilistic model known as hidden Markov model (HMM). Consider there are n states in the HMM. The particular isolated speech sig