Hidden Markov Models

Abstract. One hundred years removed from AA Markov's development of his chains, we take stock of the field he generated and the mathematical impression he left. As a tribute to Markov, we present what we consider to be the five greatest applications of Markov chains ...

An offline cursive handwriting recognition system, based on hybrid of Neural Networks (NN) and Hidden Markov Models (HMM), is described in this paper. Applying SegRec principle, the recognizer does not make hard decision at the character segmentation process. Instead, it delays the character segmentation to the recognition stage by generating a segmentation graph that describes all possible ways to segment a word into letters. To recognize a word, the NN computes the observation probabilities for each segmentation candidates (SCs) in the segmentation graph. Then, using concatenated letter-HMMs, a likelihood is computed for each word in the lexicon by multiplying the probabilities over the best paths through the graph. We present in detail two approaches to train the word recognizer: 1). character-level training 2). word-level training. The recognition performances of the two systems are discussed. I.

In recent years, there are many great successes in using deep architectures for unsupervised feature learning from data, especially for images and speech. In this paper, we introduce recent advanced deep learning models to classify two emotional categories (positive and negative) from EEG data. We train a deep belief network (DBN) with differential entropy features extracted from multichannel EEG as input. A hidden markov model (HMM) is integrated to accurately capture a more reliable emotional stage switching. We also compare the performance of the deep models to KNN, SVM and Graph regularized Extreme Learning Machine (GELM). The average accuracies of DBN-HMM, DBN, GELM, SVM, and KNN in our experiments are 87.62%, 86.91%, 85.67%, 84.08%, and 69.66%, respectively. Our experimental results show that the DBN and DBN-HMM models improve the accuracy of EEGbased emotion classification in comparison with the state-ofthe-art methods.

In this paper, we formulate a coupled factorial hidden Markov model-based framework to diagnose dependent faults occurring over time. In our previous research [1][2], the problem of diagnosing dynamic multiple faults (DMFD) is solved by assuming that the faults are independent. Here, we extend this formulation to determine the most likely evolution of dependent fault states (NP-hard problem), the one that best explains the observed test outcomes over time. An iterative Gauss-Seidel coordinate ascent optimization method along with the coupling assumptions (mixed memory Markov model) is proposed for solving the dynamic coupled fault diagnosis (DCFD) problem. A soft Viterbi algorithm is also implemented within the framework for decoding dependent fault states over time. We demonstrate the algorithm on small-scale and real-world systems and the simulation results show that this approach improves the correct isolation rate as compared to the formulation with independent fault states (DMFD). 12

Recognizing human action from image sequences is an active area of research in computer vision. In this paper, we present a novel method for human action recognition from image sequences in different viewing angles that uses the Cartesian component of optical flow velocity and human body shape feature vector information. We use principal component analysis to reduce the higher dimensional shape feature space into low dimensional shape feature space. We represent each action using a set of multidimensional discrete hidden Markov model and model each action for any viewing direction. We performed experiments of the proposed method by using KU gesture database. Experimental results based on this database of different actions show that our method is robust.

We present two real-time hidden Markov model-based systems for recognizing sentence-level continuous American Sign Language (ASL) using a single camera to track the user's unadorned hands. The first system observes the user from a desk mounted camera and achieves 92% word accuracy. The second system mounts the camera in a cap worn by the user and achieves 98% accuracy (97% with an unrestricted grammar). Both experiments use a 40 word lexicon.

—Face recognition has become an important subject in modern life, especially in security and surveillance applications. This work introduces a face recognition method, which is characterized by high-speed, low-complexity, and high-efficiency in a noisy environment. The performance of this method is greatly improved by using a median filter, such that each image is filtered to eliminate the influence of noise and light illumination. Multiple levels of discrete wavelet transform are applied to the filtered image to reduce size and eliminate further noise. Subsequently, the resultant image is scanned using a window with a predefined overlap in raster fashion to construct a sequence of observation vectors used as the basis of a model. The model consists of two states of a continuous hidden Markov model, a unique model in the face recognition field that has not yet been used by other researchers, which interprets the novelty and low complexity of the method. In spite of the presence of 0.15, 0.4, and 0.25 measurement values of impulse noise density in images stored in the ORL, Yale, and EURECOM Kinect face databases, respectively, the proposed work has achieved a recognition rate of 100%. I. INTRODUCTION Face recognition has recently received considerable attention due to its important applications in many areas and fields such as credit card verification, security, and surveillance. The advantage of face recognition over other biometric techniques is the mechanism to recognize persons, which is frequently performed without requiring that the person have physical contact with the device, e.g. use of palm or fingers. The face image can be captured by a camera without informing the target person, especially in the case of criminal surveillance and apprehension. The methods introduced to recognize faces vary relative to the procedure used for feature extraction and type of classifier used. The artificial neural network [1], principle component analysis (PCA) [2], independent component analysis [3], linear discriminant analysis (LDA) [4], support vector machine [5] and K-nearest neighbour [6] are the most widely used methods. In addition, the hidden Markov model (HMM) [7] has been successfully used in face recognition during the last two decades. The advantage of applying HMM to face recognition is the flexibility of the selection of training models, such that it is easy to add or remove individuals. There is no need for retraining the overall system; however, a model with updated images must be retrained. For dimensionality reduction, a particular feature extraction method is applied to form a sequence of observation vectors to model faces. Thereafter, for each individual, one HMM model is constructed by training a specified set of images for that individual. The training process depends on the estimation of the HMM parameters [8], such that the process is iterated several times before it converges. In the testing process, the algorithm computes the probability of the observations of unknown face image calibrated with the parameters derived from the training. The unknown face belongs to the person whose model computes the highest probability. The disadvantages of using more states of the HMM are: high computational complexities, large memory requirements, and low processing speed. The proposed work addresses the unfavourable features by using only two states of the HMM.

In this paper, we present a mismatch-aware stochastic matching (MASM) algorithm to alleviate the performance degradation under mismatched training and testing conditions. MASM first computes a reliability measure of applying a set of pre-trained speech models to a mismatch test utterance along the time axis or among different feature vector components. It then estimates and compensates the mismatch using the reliability measure to guide the speech segmentation. Experiments on a serious mismatched condition with training on PSTN-speech database and testing on mobile GSM-speech database showed that MASM outperformed the stochastic match (SM) method, especially, for short utterances.

In order to automatically provide the most appropriate learning objects to e-learners, a special interest should be given to the process of building the learners' models. First, we need to identify which relevant information to include in the learner's model, while taking into account various pedagogical considerations, and second how to accurately infer the learners' preferences and characteristics from their online behavior and activities. This paper presents a preliminary study of the possibility of integrating educational preferences in the learner's model and detecting them automatically within e-learning systems.

This paper focuses on microphone arrays to realize distant-talking speech recognition in real environments. In distant-talking situations, users can speak at arbitrary positions while moving. Therefore, it is very important for high quality speech acquisition using microphone arrays to localize a talker accurately. However, it is very difficult to localize a moving talker in noisy and reverberant environments. The talker localization errors result in performance degradation of speech recognition. One way to solve this problem is to integrate the speech recognition process and the talker localization into a unified framework.

In Natural Language Processing (NLP), Word segmentation and Part-of-Speech (POS) tagging are fundamental tasks. The POS information is also necessary in NLP's preprocessing work applications such as machine translation (MT), information retrieval (IR), etc. Currently, there are many research efforts in word segmentation and POS tagging developed separately with different methods to get high performance and accuracy. For Myanmar Language, there are also separate word segmentors and POS taggers based on statistical approaches such as Neural Network (NN) and Hidden Markov Models (HMMs). But, as the Myanmar language's complex morphological structure, the OOV problem still exists. To keep away from error and improve segmentation by utilizing POS data, segmentation and labeling should be possible at the same time.The main goal of developing POS tagger for any Language is to improve accuracy of tagging and remove ambiguity in sentences due to language structure. This paper focuses on developing word segmentation and Part-of-Speech (POS) Tagger for Myanmar Language. This paper presented the comparison of separate word segmentation and POS tagging with joint word segmentation and POS tagging.

The main steps of document processing have been reviewed, especially those implemented on Arabic writing. The techniques used in this research, such as Vector Quantization (VQ), Hidden Markov Models (HMM), and Induction of Decision Trees (ID3) have been considered, as well as reviewing pre-processing and feature extraction used in Arabic writing.

This paper will discuss the application of Markov chains in DNA sequencing. Specifically, we will look into the discrete and continuous stochastic process properties that appear in the discrete state space of various nucleotide base pairings models. The focus will be two models: Jukes-Cantor (JC69), the Kimura (K80) and their extensions.

This paper presents the design of a FPGA-based hardware co-processor, based on the SPHINX 3 speech recognition engine from CMU; capable of performing Acoustic Modeling (AM) for medium sized vocabularies in real-time. By creating an input-driven pipeline for performing the ...

Diagnosis classifies the present state of operation of the equipment, and prognosis predicts the next state of operation and its remaining useful life. In this paper, a prognosis method for the gear faults in dc machines is presented. The proposed method uses the time-frequency features extracted from the motor current as machine health indicators and predicts the future state of fault severity using hidden Markov models (HMMs). Parameter training of HMMs generally needs huge historical data, which are often not available in the case of electrical machines. Methods for computing the parameters from limited data are presented. The proposed prognosis method uses matching pursuit decomposition for estimating state-transition probabilities and experimental observations for computing state-dependent observation probability distributions. The proposed method is illustrated by examples using data collected from the experimental setup.

As amounts of publicly available video data grow, the need to automatically infer semantics from raw video data becomes significant. In this paper, we address the use of three different techniques that support that task, namely, spatio-temporal rule-based method, hidden Markov models, and dynamic Bayesian networks. First, the application of these techniques for detection and recognition of diverse events is briefly described using two case studies (Tennis and Formula 1). We explain the relationships and differences of the three approaches, as well as benefits of their integrated use. Then the focus is moved to the main point of the paper, which is the integration of the aforementioned techniques within a database management system, which provides efficient, flexible, scalable, and domain independent content-based video retrieval. We identify and consider the most important issues when extending a traditional database management system with content-based video retrieval functionality, namely issues concerning video data models, dynamic feature extraction, and extensions of different layers of database architecture. The advantages of the integrated system are demonstrated on examples from our two case studies.

The ability to predict episodes of acute hypotension (abnormal drop in arterial blood pressure) would be of immense benefit to the healthcare community, and is therefore a focus of research in both medical and engineering domains. This paper presents the use of Hidden Markov Models to predict the onset of acute hypotension, using blood pressure measurements over time. Our use of HMMs has been motivated by their ability to characterize sequential/temporal trends in a given time signal. This lends the ability to infer the health status based on blood pressure information collected over an interval of time, rather than just instantaneous measurements. We have tested the proposed technique on standard physiological signal datasets available online and have obtained promising results. As part of a bigger project, we see potential in the proposed technique being used in real time health monitoring systems.

In this work, we provide a variational Bayesian (VB) treatment of multistream fused hidden Markov models (MFHMMs), and we apply it in the context of active learning-based visual workflow recognition. Contrary to training methods yielding point estimates, such as maximum likelihood or maximum a posteriori training, the VB approach provides an estimate of the posterior distribution over the MFHMM parameters. As a result, our approach provides an elegant solution towards the amelioration of the overfitting issues of point estimate-based methods. Additionally, it provides a measure of confidence in the accuracy of the learned model, thus allowing for the easy and cost-effective utilization of active learning in the context of MFHMMs. Two alternative active learning algorithms are considered in this paper: query by committee, which selects unlabeled data that minimize the classification variance, and a maximum information gain method which aims to maximize the alteration in model variance by proper data labeling. We demonstrate the efficacy of the proposed treatment of MFHMMs by examining two challenging workflow recognition scenarios, and we show that the application of active learning, which is facilitated by our VB approach, allows for a significant reduction of the MFHMM training costs.

The basic theory of Markov chains has been known to mathematicians and engineers for close to 80 years, but it is only in the past decade that it has been applied explicitly to problems in speech processing. One of the major reasons why speech models, based on Markov chains, have not been developed until recently was the lack of a method for optimizing the parameters of the Markov model to match observed signal patterns. Such a method was proposed in the late 1960's and was immediately applied to speech processing in several research institutions. Continued refinements in the theory and implementation of Markov modelling techniques have greatly enhanced the method, leading to a wide range of applications of these models. It is the purpose of this tutorial paper to give an introduction to the theory of Markov models, and to illustrate how they have been applied to problems in speech recognition.

Online auction, shopping, electronic billing etc. all such types of application involves problems of fraudulent transactions. Online fraud occurrence and its detection is one of the challenging fields for web development and online phantom transaction. As no-secure specification of online frauds is in research database, so the techniques to evaluate and stop them are also in study. We are providing an approach with Hidden Markov Model (HMM) and mobile implicit authentication to find whether the user interacting online is a fraud or not. We propose a model based on these approaches to counter the occurred fraud and prevent the loss of the customer. Our technique is more parameterized than traditional approaches and so,chances of detecting legitimate user as a fraud will reduce.

The subspace Gaussian mixture model (SGMM) has been exploited for cross-lingual speech recognition. The general motivation is that the subspace parameters can be estimated on multiple source languages and then transferred to the target language. In this work, we investigate an extension to SGMM, referred to as subspace mixture model (SMM), in which subspace parameters on the target language are casted as a linear mixture of the subspaces derived from source languages. This approach reduces the number of SGMM model parameters, while retaining the flexibility of subspace learning on the target language. Experiments show that the proposed SMM method outperforms SGMM significantly when the target language has limited training data.

One of the research goals in the human-computer interaction community is to build believable Embodied Conversational Agents, that is, agents able to communicate complex information with human-like expressiveness and naturalness. Since emotions play a crucial role in human communication and most of them are expressed through the face, having more believable ECAs implies to give them the ability of displaying emotional facial expressions.

Hidden Markov models have been applied in many different fields, including econometrics and finance. However, the lion's share of the investigated models concerns Markovian mixtures of Gaussian distributions. We present an extension to conditional t-distributions, including models with unequal distribution types in different states. It is shown that the extended models, on the one hand, reproduce various stylized facts of daily returns better than the common Gaussian model. On the other hand, robustness to outliers and persistence of the visited states increases significantly.

Abstract. In RoboCup Middle Size league (MSL) the main referee uses assisting technology, controlled by a second referee, to support him, in particular for conveying referee decisions for robot players with the help of a wireless communication system. In this paper a vision-based system is introduced, able to interpret dynamic and static gestures of the referee, thus eliminating the need for a second one. The referee's ges-tures are interpreted by the system and sent directly to the Referee Box, which sends the proper commands to the robots. The system is divided into four modules: a real time hand tracking and feature extraction, a SVM (Support Vector Machine) for static hand posture identification, an HMM (Hidden Markov Model) for dynamic unistroke hand gesture recognition, and a FSM (Finite State Machine) to control the various system states transitions. The experimental results showed that the system works very reliably, being able to recognize the combination of gestures a...

This paper describes a new identity authentication technique by a synergetic use of lip-motion and speech. The lip-motion is defined as the distribution of apparent velocities in the movement of brightness patterns in an image and is estimated by computing the velocity components of the structure tensor by 1D processing, in 2D manifolds. Since the velocities are computed without extracting the speaker's lip-contours, more robust visual features can be obtained in comparison to motion features extracted from lip-contours. The motion estimations are performed in a rectangular lip-region, which affords increased computational efficiency. A person authentication implementation based on lip-movements and speech is presented along with experiments exhibiting a recognition rate of 98%. Besides its value in authentication, the technique can be used naturally to evaluate the ''liveness'' of someone speaking as it can be used in text-prompted dialogue. The XM2VTS database was used for performance quantification as it is currently the largest publicly available database (%300 persons) containing both lip-motion and speech. Comparisons with other techniques are presented.

Verbmobil, a German research project, aims at machine translation of spontaneous speech input. The ultimate goal is the development of a portable machine translator that will allow people to negotiate in their native language. Within this project the University of Karlsruhe has developed a speech recognition engine that has been evaluated on a yearly basis during the project and shows very promising speech recognition word accuracy results on large vocabulary spontaneous speech. We introduce the Janus Speech Recognition Toolkit underlying the speech recognizer. The main new contributions to the acoustic modeling part of our 1996 evaluation system-speaker normalization, channel normalization and polyphonic clustering-are discussed and evaluated. Besides the acoustic models we delineate the different language models used in our evaluation system: word trigram models interpolated with class based models and a separate spelling language model were applied. As a result of using the toolk...

Cel pracy: 1) zbadanie wpływu różnorodnych czynników na skuteczność i szybkość rozpoznawania izolowanych słów mowy polskiej; 2) opracowanie efektywnej metody rozpoznawania izolowanych słów polskich. Zbadano wpływ częstotliwości próbkowania sygnału mowy oraz liczby poziomów kwantyzacji na efektywność niezależnego i zależnego od mówcy rozpoznawania izolowanych wypowiedzi polskich z bazy CORPORA za pomocą metody nieliniowej transformacji czasowej (DTW) oraz niejawnych modeli Markowa HMM. Wykazano istnienie optimum doboru częstotliwości próbkowania oraz liczby poziomów kwantyzacji. Zbadano również częściowo zależność rozpoznawania izolowanych wypowiedzi od płci mówcy, wielkości zbioru uczącego oraz wybranych parametrów niejawnych modeli Markowa. Podniesiono skuteczność rozpoznawania izolowanych wypowiedzi polskich za pomocą metody nieliniowej transformacji czasowej poprzez zastosowanie wstępnej klasyfikacji za pomocą: 1) znaczników fonetycznych; 2) niejawnych modeli Markowa.

The battery management system (BMS) is an integral part of an automobile. It protects the battery from damage, predicts battery life, and maintains the battery in an operational condition. The BMS performs these tasks by integrating one or more of the functions, such as protecting the cell, thermal management, controlling the charge-discharge, determining the state of charge (SOC), state of health (SOH), and remaining useful life (RUL) of the battery, cell balancing, data acquisition, communication with on-board and off-board modules, as well as monitoring and storing historical data. In this paper, we propose a BMS that estimates the critical characteristics of the battery (such as SOC, SOH, and RUL) using a data-driven approach. Our estimation procedure is based on a modified Randles circuit model consisting of resistors, a capacitor, the Warburg impedance for electrochemical impedance spectroscopy test data, and a lumped parameter model for hybrid pulse power characterization test data. The resistors in a Randles circuit model usually characterize the self-discharge and internal resistance of the battery, the capacitor generally represents the charge stored in the battery, and the Warburg impedance represents the diffusion phenomenon. The Randles circuit parameters are estimated using a frequency-selective nonlinear least squares estimation technique, while the lumped parameter model parameters are estimated by the prediction error minimization method. We investigate the use of support vector machines (SVMs) to predict the capacity fade and power fade, which characterize the SOH of a battery, as well as estimate the SOC of the battery. An alternate procedure for estimating the power fade and energy fade from low-current Hybrid Pulse Power characterization (L-HPPC) test data using the lumped parameter battery model has been proposed. Predictions of RUL of the battery are obtained by support vector regression of the power fade and capacity fade estimates. Survival function estimates for reliability analysis of the battery are obtained using a hidden Markov model (HMM) trained using time-dependent estimates of capacity fade and power fade as observations. The proposed framework provides a systematic way for estimating relevant battery characteristics with a high-degree of accuracy.

This paper describes an effect of articulatory dynamic parameters (Δ and ΔΔ) on neural network based automatic speech recognition(ASR). Articulatory features (AFs) or distinctive phonetic features (DPFs)-based system shows its superiority in performances over acoustic features-based in ASR. These performances can be further improved by incorporating articulatory dynamic parameters into it. In this paper, we have proposed such a phoneme recognition system that comprises three stages: (i) DPFs extraction using a multilayer neural network (MLN) from acoustic features, (ii) incorporation of dynamic parameters into another MLN for reducing DPF context, and (iii) addition of an Inhibition/Enhancement (In/En) network for categorizing the DPF movement more accurately and Gram-Schmidt (GS) orthogonalization procedure for decorrelating the inhibited/enhanced data vector before connecting with hidden Markov model (HMMs)-based classifier. From the experiments on Japanese Newspaper Article Sentences (JNAS), it is observed that the proposed method provides a higher phoneme correct rate over the method that does not incorporate dynamic articulatory parameters. Moreover, it reduces mixture components in HMM for obtaining a higher recognition performance.

A crucial part of a speech recognizer is the acoustic feature extraction, especially when the application is intended to be used in noisy environment. In this paper we investigate several novel front-end techniques and compare them to multiple baselines. Recognition tests were performed on studio quality wide band recordings on Hungarian as well as on narrow band telephone speech including real-life noises collected in six languages: English, German, French, Italian, Spanish and Hungarian. The following baseline feature types were used with several settings: Mel Frequency Cepstral Coefficients (MFCC),

We propose a novel multi-stream framework for continuous conversational speech recognition which employs bidirectional Long Short-Term Memory (BLSTM) networks for phoneme prediction. The BLSTM architecture allows recurrent neural nets to model longrange context, which led to improved ASR performance when combined with conventional triphone modeling in a Tandem system. In this paper, we extend the principle of joint BLSTM and triphone modeling to a multi-stream system which uses MFCC features and BLSTM predictions as observations originating from two independent data streams. Using the COSINE database, we show that this technique prevails over a recently proposed single-stream Tandem system as well as over a conventional HMM recognizer.

Usage monitoring entails determining the actual usage of a component on the aircraft and requires accurate recognition of regimes. In this paper, a data mining approach is adopted for regime recognition. In particular, a regime recognition algorithm developed based on hidden Markov models is presented. The developed algorithm was validated using the flight card data of an Army UH-60L helicopter. The performance of this regime recognition algorithm was also compared with other data mining methods using the same dataset. Using the flight card information and regime definitions, a dataset of about 56,000 data points labeled with 50 regimes recorded in the flight card were mapped to the health and usage monitoring parameters. The validation and performance comparison results have showed that the hidden Markov model based regime recognition algorithm was able to accurately recognize the regimes recorded in the flight card data and outperformed other data mining methods.

livre reconnaisance et apprentissage des formes et du son : b preface, shrikanth narayanan and abeer alwan (editors). /bb foreword:modification of speech: tribute to mike macon, jan van santen. ... Sommaire. Preface, Shrikanth Narayanan and Abeer Alwan (Editors). ...

Traffic congestion is an important socioeconomic problem that swelled in the last few decades. It affects the social mobility of people, length of trips, quality of life, and the economy of countries. As a major problem in most countries, it has been tackled by governments, universities, and advanced research using intelligent transportation systems (ITS) to solve the problem or at least ease its adverse effects. Hidden Markov Models (HMM) represent one of the methods that are suitable for congestion prediction. In this paper, a new model, based on Hidden Markov Model and Contrast, is proposed to define the traffic states during peak hours in two dimensional space (2D). The proposed model uses mean speed and contrast to capture the variability in traffic patterns. Empirical evaluation shows that the proposed approach has improved prediction error in comparison to HMM related work and neuro-fuzzy approaches. Ó 2019 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/ by-nc-nd/4.0/).

Current advances in technology allow for the efficient capturing and storage of high resolution and high frequency person movement data. The advent of wi-fi position triangulation has allowed us to capture human movement with a great deal of accuracy inside a closed urban structure, e.g., a University or a shopping mall and at a very high frequency (an observation every second). While there have been significant advances in our ability to capture this data, advances in robust modelling techniques for movement prediction have been largely absent. In this paper we aim to present our theoretical insights based on person movement data collected from Deloitte University, West Lake Texas. We outline two independent approaches -the first aims to model the problem of person movement prediction while the second aims to classify people based on their movement history. This paper presents a technique to model person movement data using a Hidden Markov Model on a large data set with multiple rooms and locations of importance as well as a technique to classify users based on their path data (movement traces). We achieve an accuracy of 65% in the predictions of the discrete Hidden Markov Model.