Novel Extended Phonemic Set for Mandarin Continuous Speech Recognition

An extended phonemic set of mandarin from the view of speech recognition is proposed. This set absorbs most principles of some other existing phonemic sets for mandarin, like Worldbet and SAMPA-C, and also takes advantage of some practical experiences from speech recognition research for increasing the discriminability between word models. And the experiments in speaker independent continuous speech recognition show that hidden Markov models defined by this phonemic set have a better performance than those based on initial/final units of mandarin and have a very compact size.     

Linear Discriminant Analysis and Kernel Vector Quantization for Mandarin Digits Recognition

Linear discriminant analysis and kernel vector quantization are integrated into vector quantization based speech recognition system for improving the recognition accuracy of Mandarin digits. These techniques increase the class separability and optimize the clustering procedure. Speaker-dependent (SD) and speaker-independent (SI) experiments are performed to evaluate the performance of the proposed method. The experiment results show that the proposed method is capable of reaching the word error rate of 3.76 % in SD case and 6.60 % in SI case. Such a system can be suitable for being embedded in personal digital assistant(PDA), mobile phone and so on to perform voice controlling such as digit dialing, calculating, etc.     

SVM with discriminative dynamic time alignment

In the past several years, support vector machines (SVM) have achieved a huge success in many fields, especially in pattern recognition. But the standard SVM cannot deal with length-variable vectors, which is one severe obstacle for its applications to some important areas, such as speech recognition and part-of-speech tagging. The paper proposed a novel SVM with discriminative dynamic time alignment (DDTA-SVM) to solve this problem. When training DDTA-SVM classifier, according to the category information of the training samples, different time alignment strategies were adopted to manipulate them in the kernel functions, which contributed to great improvement for training speed and generalization capability of the classifier. Since the alignment operator was embedded in kernel functions, the training algorithms of standard SVM were still compatible in DDTA-SVM. In order to increase the reliability of the classification, a new classification algorithm was suggested. The preliminary experimental results on Chinese confusable syllables speech classification task show that DDTA-SVM obtains faster convergence speed and better classification performance than dynamic time alignment kernel SVM (DTAK-SVM). Moreover, DDTA-SVM also gives higher classification precision compared to the conventional HMM. This proves that the proposed method is effective, especially for confusable length-variable pattern classification tasks.     

Automatic target recognition of moving target based on empirical mode decomposition and genetic algorithm support vector machine

In order to improve measurement accuracy of moving target signals, an automatic target recognition model of moving target signals was established based on empirical mode decomposition(EMD) and support vector machine(SVM). Automatic target recognition process on the nonlinear and non-stationary of Doppler signals of military target by using automatic target recognition model can be expressed as follows. Firstly, the nonlinearity and non-stationary of Doppler signals were decomposed into a set of intrinsic mode functions(IMFs) using EMD. After the Hilbert transform of IMF, the energy ratio of each IMF to the total IMFs can be extracted as the features of military target. Then, the SVM was trained through using the energy ratio to classify the military targets, and genetic algorithm(GA) was used to optimize SVM parameters in the solution space. The experimental results show that this algorithm can achieve the recognition accuracies of 86.15%, 87.93%, and 82.28% for tank, vehicle and soldier, respectively.     

Leaching of a copper flotation concentrate with ammonium persulfate in an autoclave system

The leaching behavior of a copper flotation concentrate was investigated using ammonium persulfate （APS） in an autoclave systee. The decomposition products of APS, active oxygen, and acidic medium were used to extract metals from the concentrate. Leaching experiments were performed to compare the availability of APS as an oxidizing agent for leaching of the concentrate under atmospheric conditions and in an autoclave system. Leaching temperature and APS concentration were found to be important parameters in both leaching systems. Atmospheric leaching studies showed that the metal extractions increased with the increase in APS concentration and temperature （up to 333 K）. A similar tendency was determined in the autoclave studies up to 423 K. It was also determined that the metal extractions decreased at temperatures above 423 K due to the passivation of the particle surface by molten elemental sulfur. The results showed that higher copper extractions could be achieved using an autoclave system.     

Application of particle trajectory model in 1D planar ejection

A simple one-dimensional planar model for ejection was set up based on experiments.And numerical simulation was performed on this model with particle trajectory model method.An Eulerian finite volume method was conducted to resolve gas field.And Lagrangian method was imposed to track each particle.The interaction between gas and particles was responded as source terms in governing equations which were induced by forces.The effects of total spraying mass,particle size and other factors on the mixture of particles and gas were investigated.The spatial distributions of particle mass and velocity at different time were presented.The result shows that the numerical results are qualitatively consistent to those of experiments.     

Comparative Study on Channel Compensation for Robust Speech Recognition

Some channel compensation techniques integrated into front-end of speech recognizer for improving channel robustness are described. These techniques include cepstral mean normalization, rasta processing and blind equalization. Two standard channel frequency characteristics, G. 712 and MIRS, are introduced as channel distortion references and a mandarin digit string recognition task is performed for evaluating and comparing the performance of these different methods. The recognition results show that in G. 712 case blind equalization can achieve the best recognition performance while cepstral mean normalization outperforms the other methods in MIRS case which is capable of reaching a word error rate of 3.96 %.     

Nitrogen removal from municipal wastewater by limit-oxic/anoxic/oxic biological aerated filter system

In this study,a three-stage biological aerated filter(BAF) system was proposed for the enhancement of nitrogen removal in the treatment of low carbon-to-nitrogen ratio(C/N ratio) municipal wastewater.Operational parameters were studied for each process for maximum nitrite accumulation in the nitrification step and nitrite adaptation in the denitrification step.Nitrite accumulation during nitrification could be controlled by the dissolved oxygen(DO) concentration,presenting a mean value of 40% at around 1.0 mg DO/L.Denitrification could be adapted to nitrite and the process was stable if nitrite in the reactor was keep low.Once the operational parameters were established,the process was stable and a steady state was maintained for over 30 days,and the various indexes of discharged water were up to the Discharge standard of pollutants for municipal wastewater treatment plant(GB18918-2002) Level-one A.It was concluded that the three-stage BAF system proposed in this study was excellent in nitrogen removal performance by employing three-column functioning as short-cut nitrification,short-cut denitrification and secondary nitrification,respectively.     

Finite element and experimental analysis for cold tube-bending of a power station boiler

Combining the computer simulation, theoretical analysis and experiments, this paper studies the process of tube-bending production of a power station boiler. The FEM-numerical simulation system was set up to study the performance features in the process of manufacturing cold tube-bending, such as ovality, thinning ratio of wall and the equipment driven force. Experiments were carried out in Harbin boiler factory. The parametric FEM model can simulate the force and quality parameters for tube-bending process and the error is within 5%. Experiments have demonstrated that the results achieved by this system are more useful and economical than those obtained through trial-produce method. It greatly reduces the production cost using computer simulation than conducting lots of experiments.     

基于VQ的高效汉语语音识别方法的比较性研究

介绍了一种基于VQ的小词汇量特定人快速语音识别方法,并对该方法中的关键参数进行了比较性研究. 这一方法尤其适用于汉语特定人小词汇量语音识别. 与典型的HMM和NN识别算法相比,该方法复杂度低、系统资源消耗少而识别率高,在对汉语11个数字的大规模识别测试中误识率仅为3.86%. 因此,该方法适合于在手机、PDA等资源有限的系统中实现语音数字拨号、人名拨号及查询等语音控制功能.     

非特定人语音指令识别系统的研究与实现

为了研究语音识别技术的实用化问题.采用离散隐马尔可夫模型方法,基于单片微控制器加数字信号处理器(MCU DSP)构架,实现了一个非特定人语音指令识别系统.该系统适应性强,每个指令识别时间在1s内,准确率达到96%.     

Improved hidden Markov model for speech recognition and POS tagging

In order to overcome defects of the classical hidden Markov model (HMM), Markov family model (MFM), a new statistical model was proposed. Markov family model was applied to speech recognition and natural language processing. The speaker independently continuous speech recognition experiments and the part-of-speech tagging experiments show that Markov family model has higher performance than hidden Markov model. The precision is enhanced from 94.642% to 96.214% in the part-of-speech tagging experiments, and the work rate is reduced by 11.9% in the speech recognition experiments with respect to HMM baseline system.     

基于核K-均值聚类和支持向量机结合的说话人识别方法

提出了基于核K-均值聚类方法与支持向量机结合的说话人识别方法,为每两个人建立一个支持向量机,对支持向量机输入的语音信号先进行核K-均值聚类,并选取有效样本作为支持向量机的输入,本文提出的聚类方法能够去更好的聚类并约简数据,提高了识别率.实验比较了在用支持向量机作为分类器的情况下,该核聚类与传统聚类方法的训练速度和识别性能,验证了本文提出方法的有效性.     

基于特征值混和提升语音识别鲁棒性技术(英文)

语音识别技术可以为要求双手同时作业的操作人员和残疾人提供一种便捷的控制方法。作者在文中提出了一种通过结合二阶频率滤波和RASTA技术来增强语音识别鲁棒性的方法,并将这种方法成功应用于机器人化护理床的控制系统中,增强了识别系统在医院、工厂等非稳定噪声环境下语音识别的鲁棒性。通过将HMM/GMM混合模型的传统Mel频率倒谱系数为特征值的识别系统与HMM/GMM混合模型的RASTA-FF2为特征值的识别系统进行比较,并分别在纯语音和带噪语音条件下进行测试,发现经过二阶频率滤波后的FF2特征值再经过RASTA滤波器滤波,特别是在非稳定噪声环境下,以RASTA-FF2为特征值的识别系统比传统的识别系统的识别率更高,这表明FF2特征值与RASTA滤波器技术相结合,一个作用于频域,一个作用于时间域,可以有效地消除语音信号中的不同噪声成份。     

基于非齐次隐马尔可夫模型的特定人元音的识别方法

针对特定人汉语元音的语音识别,提出一种基于非齐次隐马尔可夫模型的识别方法.该方法首先提取声道频率响应作为特征参数,然后建立非齐次隐马尔可夫模型来更为精确地刻画真实的语音现象,接着进行语音识别实验,并与齐次隐马尔可夫模型进行比较.实验结果表明该方法可以使特定人的元音的识别率达到98.73%,明显改变了识别系统的性能.该方法具有很好的理论研究前景和实际应用价值.     

双模型语音识别中的听视觉合成和模型同步异步性实验研究

研究了双模型语音识别系统中前合成和后合成两种听觉视觉合成方法；同时在后合成方法中引入了考虑听觉和视觉同步异步特点的复合模型。仿真实验证明了在声学噪音环境下，后合成方法能够带来比较理想的识别效果；考虑听觉和视觉同步异步性的模型可以有效地提高识别率。