基于HMM模型的语音单元边界的自动切分

基于隐尔马可夫模型(HMM)的强制对齐方法被用于文语转换系统(TTS)语音单元边界切分。为提高切分准确性,本文对HMM模型的特征选择,模型参数和模型聚类进行优化。实验表明:12维静态M e l频率倒谱系数(M FCC)是最优的语音特征;HMM模型中的状态模型采用单高斯;对于特定说话人的HMM模型,使用分类与衰退树(CART)聚类生成的绑定状态模型个数在3 000左右最优。在英文语音库中音素边界切分的实验中,切分准确率从模型优化前的77.3%提高到85.4%。

Ontology mapping based on hidden Markov model

The existing ontology mapping methods mainly consider the structure of the ontology and the mapping precision is lower to some extent. According to statistical theory, a method which is based on the hidden Markov model is presented to establish ontology mapping. This method considers concepts as models, and attributes, relations, hierarchies, siblings and rules of the concepts as the states of the HMM, respectively. The models corresponding to the concepts are built by virtue of learning many training instances. On the basis of the best state sequence that is decided by the Viterbi algorithm and corresponding to the instance, mapping between the concepts can be established by maximum likelihood estimation. Experimental results show that this method can improve the precision of heterogeneous ontology mapping effectively.

Suppression of Speckle in SAR Images Using Wavelet-Based HMM

In order to suppress the speckle appearing in synthesis aperture radar (SAR) images, a novel speckle reduction method based on wavelet domain hidden Markov tree (HMT) was proposed. First, the image was logarithmic transformed to change the statistical property of the speckles. Then an HMT was constructed in the correspondent wavelet domain. Based on this model, the image signal was restored by maximum likelihood estimation and speckle noise was suppressed. Simulating SAR images had shown that the performance of the filter is satisfactory for both speckle smoothing and edges presentation, and for generating visually natural images as well.

Frame erasure concealment in wideband speech coding based on large hidden Markov model

Frame erasure concealment is studied to solve the problem of rapid speech quality reduction due to the loss of speech parameters during speech transmission. A large hidden Markov model is applied to model the immittance spectral frequency （ISF） parameters in AMR-WB codec to optimally estimate the lost ISFs based on the minimum mean square error （MMSE） rule. The estimated ISFs are weighted with the ones of their previous neighbors to smooth the speech, resulting in the actual concealed ISF vectors. They are used instead of the lost ISFs in the speech synthesis on the receiver. Comparison is made between the speech concealed by this algorithm and by Annex I of G. 722. 2 specification, and simulation shows that the proposed concealment algorithm can lead to better performance in terms of frequency-weighted spectral distortion and signal-to-noise ratio compared to the baseline method, with an increase of 2.41 dB in signal-to-noise ratio （SNR） and a reduction of 0. 885 dB in frequency-weighted spectral distortion.

On-line Fault Diagnosis in Industrial Processes Using Variable Moving Window and Hidden Markov Model

An integrated framework is presented to represent and classify process data for on-line identifying abnormal operating conditions. It is based on pattern recognition principles and consists of a feature extraction step, by which wavelet transform and principal component analysis are used to capture the inherent characteristics from process measurements, followed by a similarity assessment step using hidden Markov model (HMM) for pattern comparison. In most previous cases, a fixed-length moving window was employed to track dynamic data, and often failed to capture enough information for each fault and sometimes even deteriorated the diagnostic performance. A variable moving window, the length of which is modified with time, is introduced in this paper and case studies on the Tennessee Eastman process illustrate the potential of the proposed method.

Pattern recognition of optimal traffic path based on HMM

In order to alleviate urban traffic congestion and provide fast vehicle paths,a hidden Markov model(HMM)based on multi-feature data of urban regional roads is constructed to solve the problems of low recognition rate and poor instability of traditional model algorithms.At first,the HHM is obtained by training.Then according to dynamic planning principle,the traffic states of intersections are obtained by the Viterbi algorithm.Finally,the optimal path is selected based on the obtained traffic states of intersections.The experiment results show that the proposed method is superior to other algorithms in road unobstruction rate and recognition rate under complex road conditions.

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.

Place recognition based on saliency for topological localization

Based on salient visual regions for mobile robot navigation in unknown environments, a new place recognition system was presented. The system uses monocular camera to acquire omni-directional images of the environment where the robot locates. Salient local regions are detected from these images using center-surround difference method, which computes opponencies of color and texture among multi-scale image spaces. And then they are organized using hidden Markov model (HMM) to form the vertex of topological map. So localization, that is place recognition in our system, can be converted to evaluation of HMM. Experimental results show that the saliency detection is immune to the changes of scale, 2D rotation and viewpoint etc. The created topological map has smaller size and a higher ratio of recognition is obtained.

Predicting Complex Word Emotions and Topics through a Hierarchical Bayesian Network

In this paper, we provide a Word Emotion Topic （WET） model to predict the complex word e- motion information from text, and discover the dis- trbution of emotions among different topics. A complex emotion is defined as the combination of one or more singular emotions from following 8 basic emotion categories： joy, love, expectation, sur- prise, anxiety, sorrow, anger and hate. We use a hi- erarchical Bayesian network to model the emotions and topics in the text. Both the complex emotions and topics are drawn from raw texts, without con- sidering any complicated language features. Our ex- periment shows promising results of word emotion prediction, which outperforms the traditional parsing methods such as the Hidden Markov Model and the Conditional Random Fields（CRFs） on raw text. We also explore the topic distribution by examining the emotion topic variation in an emotion topic diagram.

Fuzzy C-Means Clustering Based Phonetic Tied-Mixture HMM in Speech Recognition

A fuzzy clustering analysis based phonetic tied-mixture HMM(FPTM) was presented to decrease parameter size and improve robustness of parameter training. FPTM was synthesized from state-tied HMMs by a modified fuzzy C-means clustering algorithm. Each Gaussian codebook of FPTM was built from Gaussian components within the same root node in phonetic decision tree. The experimental results on large vocabulary Mandarin speech recognition show that compared with conventional phonetic tied-mixture HMM and state-tied HMM with approximately the same number of Gaussian mixtures, FPTM achieves word error rate reductions by 4.84% and 13.02% respectively. Combining the two schemes of mixing weights pruning and Gaussian centers fuzzy merging, a significantly parameter size reduction was achieved with little impact on recognition accuracy.

RESEARCH OF PINYIN-TO-CHARACTER CONVERSION BASED ON MAXIMUM ENTROPY MODEL

This paper applied Maximum Entropy (ME) model to Pinyin-To-Character (PTC) conversion in-stead of Hidden Markov Model (HMM) that could not include complicated and long-distance lexical informa-tion. Two ME models were built based on simple and complex templates respectively, and the complex one gave better conversion result. Furthermore, conversion trigger pair of y A → y B cBwas proposed to extract the long-distance constrain feature from the corpus; and then Average Mutual Information (AMI) was used to se-lect conversion trigger pair features which were added to the ME model. The experiment shows that conver-sion error of the ME with conversion trigger pairs is reduced by 4% on a small training corpus, comparing with HMM smoothed by absolute smoothing.

Hybrid SVM/HMM Method for Face Recognition

A face recognition system based on Support Vector Machine (SVM) and Hidden Markov Model (HMM) has been proposed. The powerful discriminative ability of SVM is combined with the temporal modeling ability of HMM. The output of SVM is moderated to be probability output, which replaces the Mixture of Gauss (MOG) in HMM. Wavelet transformation is used to extract observation vector, which reduces the data dimension and improves the robustness.The hybrid system is compared with pure HMM face recognition method based on ORL face database and Yale face database. Experiments results show that the hybrid method has better performance.

AN HMM BASED ANALYSIS FRAMEWORK FOR SEMANTIC VIDEO EVENTS

Semantic video analysis plays an important role in the field of machine intelligence and pattern recognition. In this paper, based on the Hidden Markov Model (HMM), a semantic recognition framework on compressed videos is proposed to analyze the video events according to six low-level features. After the detailed analysis of video events, the pattern of global motion and five features in foreground—the principal parts of videos, are employed as the observations of the Hidden Markov Model to classify events in videos. The applications of the proposed framework in some video event detections demonstrate the promising success of the proposed framework on semantic video analysis.

A lexicalized second-order-HMM for ambiguity resolution in Chinese segmentation and POS tagging

Hidden Markov Model（HMM） is a main solution to ambiguities in Chinese segmentation anti POS （part-of-speech） tagging. While most previous works tot HMM-based Chinese segmentation anti POS tagging eonsult POS informatiou in contexts, they do not utilize lexieal information which is crucial for resoMng certain morphologieal ambiguity. This paper proposes a method which incorporates lexieal information and wider context information into HMM. Model induction anti related smoothing technique are presented in detail. Experiments indicate that this technique improves the segmentation and tagging accuracy by nearly 1%.

Web-Based Information Extraction Technology

Information extraction techniques on the Web are the current research hotspot. Now many information extraction techniques based on different principles have appeared and have different capabilities. We classify the existing information extraction techniques by the principle of information extraction and analyze the methods and principles of semantic information adding, schema defining, rule expression, semantic items locating and object locating in the approaches. Based on the above survey and analysis, several open problems are discussed.

Discrete channel modelling based on genetic algorithm and simulated annealing for training hidden Markov model

Hidden Maxkov models （HMMs） have been used to model burst error sources of wireless channels. This paper proposes a hybrid method of using genetic algorithm （GA） and simulated annealing （SA） to train HMM for discrete channel modelling. The proposed method is compared with pure GA, and experimental results show that the HMMs trained by the hybrid method can better describe the error sequences due to SA＇s ability of facilitating hill-climbing at the later stage of the search. The burst error statistics of the HMMs trained by the proposed method and the corresponding error sequences are also presented to validate the proposed method.

QoS Provisioning Energy Saving Dynamic Access Policy for Overlay Cognitive Radio Networks with Hidden Markov Channels

Dynamic spectrum access policy is crucial in improving the performance of over- lay cognitive radio networks. Most of the previ- ous works on spectrum sensing and dynamic spe- ctrum access consider the sensing effective- ness and spectrum utilization as the design cri- teria, while ignoring the energy related issues and QoS constraints. In this article, we propose a QoS provisioning energy saving dynamic acc- ess policy using stochastic control theory con- sidering the time-varying characteristics of wir- eless channels because of fading and mobility. The proposed scheme determines the sensing action and selects the optimal spectrum using the corresponding power setting in each decis- ion epoch according to the channel state with the objective being to minimise both the flame error rate and energy consumption. We use the Hidden Markov Model （HMM） to model a wir- eless channel, since the channel state is not dir- ectly observable at the receiver, but is instead embedded in the received signal. The proced- ure of dynamic spectrum access is formulated as a Markov decision process which can be sol- ved using linear programming and the primal- dual index heuristic algorithm, and the obta- ined policy has an index-ability property that can be easily implemented in real systems. Sim- ulation results are presented to show the per- formance improvement caused by the propo- sed approach.

Video object's behavior analyzing based on motion history image and hidden markov model

A novel method was proposed, which extracted video object＇ s track and analyzed video object＇ s be- havior. Firstly, this method tracked the video object based on motion history image, and obtained the co- ordinate-based track sequence and orientation-based track sequence of the video object. Then the pro- posed hidden markov model （HMM） based algorithm was used to analyze the behavior of video object with the track sequence as input. Experimental results on traffic object show that this method can achieve the statistics of a mass of traffic objects＇ behavior efficiently, can acquire the reasonable velocity behavior curve of traffic object, and can recognize traffic object＇ s various behaviors accurately. It provides a base for further research on video object behavior.

An Intelligent Web Pre-fetching Based on Hidden Markov Model

Web pre-fetching is one of the most popular strategies, which are proposed for reducing the perceived access delay and improving the service quality of web server. In this paper, we present a pre-fetching model based an the hidden Markov model, which mines the later information requirement concepts that the user's access path contains and makes semantic-based pre-fetching decisions. Experimental results show that our schcme has better predictive pre-fetching precision.

Statistical Model-Based Driving Situation Recognition

The authors propose a two-stage method for recognizing driving situations on the basis of driving signals for application to a safe human interface of an in-vehicle information system. In first stage, an unknown driving situation is determined as stopping behavior or non-stopping behavior. In second stage, a Hidden Markov Model （HMM）-based pattern recognition method is used to model and recognize six non-stopping driving situations. The authors attempt to find the optimal HMM configuration to improve the performance of driving situation recognition. Center for Integrated Acoustic Information Research （CLAIR） in-vehicle corpus is used to evaluate the HMM-based recognition method. Driving situation categories are recognized using five driving signals. The proposed method achieves a relative error reduction rate of 30.9% compared to a conventional one-stage based HMMs.