Modeling patterns in temporal data has arisen as an important problem in engineering and science. This has led to the popularity of several dynamic models, in particular the renowned hidden Markov model (HMM) [Rabiner, 1989]. Despite its widespread success in many cases, the standard HMM often fails to model more complex data whose elements are correlated hierarchically or over a long period. Such problems are, however, frequently encountered in practice. Existing efforts to overcome this weakness often address either one of these two aspects separately, mainly due to computational intractability. Motivated by this modeling challenge in many real world problems, in particular, for video surveillance and segmentation, this thesis aims to develop tractable probabilistic models that can jointly model duration and hierarchical information in a unified framework. We believe that jointly exploiting statistical strength from both properties will lead to more accurate and robust models for the needed task. To tackle the modeling aspect, we base our work on an intersection between dynamic graphical models and statistics of lifetime modeling. Realizing that the key bottleneck found in the existing works lies in the choice of the distribution for a state, we have successfully integrated the discrete Coxian distribution [Cox, 1955], a special class of phase-type distributions, into the HMM to form a novel and powerful stochastic model termed as the Coxian Hidden Semi-Markov Model (CxHSMM). We show that this model can still be expressed as a dynamic Bayesian network, and inference and learning can be derived analytically. / Most importantly, it has four superior features over existing semi-Markov modelling: the parameter space is compact, computation is fast (almost the same as the HMM), close-formed estimation can be derived, and the Coxian is flexible enough to approximate a large class of distributions. Next, we exploit hierarchical decomposition in the data by borrowing analogy from the hierarchical hidden Markov model in [Fine et al., 1998, Bui et al., 2004] and introduce a new type of shallow structured graphical model that combines both duration and hierarchical modelling into a unified framework, termed the Coxian Switching Hidden Semi-Markov Models (CxSHSMM). The top layer is a Markov sequence of switching variables, while the bottom layer is a sequence of concatenated CxHSMMs whose parameters are determined by the switching variable at the top. Again, we provide a thorough analysis along with inference and learning machinery. We also show that semi-Markov models with arbitrary depth structure can easily be developed. In all cases we further address two practical issues: missing observations to unstable tracking and the use of partially labelled data to improve training accuracy. Motivated by real-world problems, our application contribution is a framework to recognize complex activities of daily livings (ADLs) and detect anomalies to provide better intelligent caring services for the elderly. / Coarser activities with self duration distributions are represented using the CxHSMM. Complex activities are made of a sequence of coarser activities and represented at the top level in the CxSHSMM. Intensive experiments are conducted to evaluate our solutions against existing methods. In many cases, the superiority of the joint modeling and the Coxian parameterization over traditional methods is confirmed. The robustness of our proposed models is further demonstrated in a series of more challenging experiments, in which the tracking is often lost and activities considerably overlap. Our final contribution is an application of the switching Coxian model to segment education-oriented videos into coherent topical units. Our results again demonstrate such segmentation processes can benefit greatly from the joint modeling of duration and hierarchy.

Aufgrund seiner nahezu kontinuierlichen eruptiven Aktivität zählt der Merapi zu den gefährlichsten Vulkanen der Welt. Der Merapi befindet sich im Zentralteil der dicht bevölkerten Insel Java (Indonesien). Selbst kleinere Ausbrüche des Merapi stellen deswegen eine große Gefahr für die ansässige Bevölkerung in der Umgebung des Vulkans dar. Die am Merapi beobachtete enge Korrelation zwischen seismischer und vulkanischer Aktivität erlaubt es, mit Hilfe der Überwachung der seismischen Aktivität Veränderungen des Aktivitätszustandes des Merapi zu erkennen. Ein System zur automatischen Detektion und Klassifizierung seismischer Ereignisse liefert einen wichtigen Beitrag für die schnelle Analyse der seismischen Aktivität. Im Falle eines bevorstehenden Ausbruchszyklus bedeutet dies ein wichtiges Hilfsmittel für die vor Ort ansässigen Wissenschaftler.<br /> In der vorliegenden Arbeit wird ein Mustererkennungsverfahren verwendet, um die Detektion und Klassifizierung seismischer Signale vulkanischen Urprunges aus den kontinuierlich aufgezeichneten Daten in Echtzeit zu bewerkstelligen. Der hier verwendete A nsatz der hidden Markov Modelle (HMM) wird motiviert durch die große Ähnlichkeit von seismischen Signalen vulkanischen Ursprunges und Sprachaufzeichnungen und den großen Erfolg, den HMM-basierte Erkennungssysteme in der automatischen Spracherkennung erlangt haben. <br /> Für eine erfolgreiche Implementierung eines Mustererkennungssytems ist es notwendig, eine geeignete Parametrisierung der Rohdaten vorzunehmen. Basierend auf den Erfahrungswerten seismologischer Observatorien wird ein Vorgehen zur Parametrisierung des seismischen Wellenfeldes auf Grundlage von robusten Analyseverfahren vorgeschlagen. Die Wellenfeldparameter werden pro Zeitschritt in einen reell-wertigen Mustervektor zusammengefasst. Die aus diesen Mustervektoren gebildete Zeitreihe ist dann Gegenstand des HMM-basierten Erkennungssystems. Um diskrete hidden Markov Modelle (DHMM) verwenden zu können, werden die Mustervektoren durch eine lineare Transformation und nachgeschaltete Vektor Quantisierung in eine diskrete Symbolsequenz überführt. Als Klassifikator kommt eine Maximum-Likelihood Testfunktion zwischen dieser Sequenz und den, in einem überwachten Lernverfahren trainierten, DHMMs zum Einsatz.<br /> Die am Merapi kontinuierlich aufgezeichneten seismischen Daten im Zeitraum vom 01.07. und 05.07.1998 sind besonders für einen Test dieses Klassifikationssystems geeignet. In dieser Zeit zeigte der Merapi einen rapiden Anstieg der Seismizität kurz bevor dem Auftreten zweier Eruptionen am 10.07. und 19.07.1998. Drei der bekannten, vom Vulkanologischen Dienst in Indonesien beschriebenen, seimischen Signalklassen konnten in diesem Zeitraum beobachtet werden. Es handelt sich hierbei um flache vulkanisch-tektonische Beben (VTB, h < 2.5 km), um sogenannte MP-Ereignisse, die in direktem Zusammenhang mit dem Wachstum des aktiven Lavadoms gebracht werden, und um seismische Ereignisse, die durch Gesteinslawinen erzeugt werden (lokaler Name: Guguran).<br /> Die spezielle Geometrie des digitalen seismischen Netzwerkes am Merapi besteht aus einer Kombination von drei Mini-Arrays an den Flanken des Merapi. Für die Parametrisierung des Wellenfeldes werden deswegen seismische Array-Verfahren eingesetzt. Die individuellen Wellenfeld Parameter wurden hinsichtlich ihrer Relevanz für den Klassifikationsprozess detailliert analysiert. Für jede der drei Signalklassen wurde ein Satz von DHMMs trainiert. Zusätzlich wurden als Ausschlussklassen noch zwei Gruppen von Noise-Modellen unterschieden.<br /> Insgesamt konnte mit diesem Ansatz eine Erkennungsrate von 67 % erreicht werden. Im Mittel erzeugte das automatische Klassifizierungssystem 41 Fehlalarme pro Tag und Klasse. Die Güte der Klassifikationsergebnisse zeigt starke Variationen zwischen den individuellen Signalklassen. Flache vulkanisch-tektonische Beben (VTB) zeigen sehr ausgeprägte Wellenfeldeigenschaften und, zumindest im untersuchten Zeitraum, sehr stabile Zeitmuster der individuellen Wellenfeldparameter. Das DHMM-basierte Klassifizierungssystem erlaubte für diesen Ereignistyp nahezu 89% richtige Entscheidungen und erzeugte im Mittel 2 Fehlalarme pro Tag.<br /> Ereignisse der Klassen MP und Guguran sind mit dem automatischen System schwieriger zu erkennen. 64% aller MP-Ereignisse und 74% aller Guguran-Ereignisse wurden korrekt erkannt. Im Mittel kam es bei MP-Ereignissen zu 87 Fehlalarmen und bei Guguran Ereignissen zu 33 Fehlalarmen pro Tag. Eine Vielzahl der Fehlalarme und nicht detektierten Ereignisse entstehen jedoch durch eine Verwechslung dieser beiden Signalklassen im automatischen Erkennnungsprozess. Dieses Ergebnis konnte aufgrund der ähnlichen Wellenfeldeigenschaften beider Signalklassen erklärt werden, deren Ursache vermutlich in den bekannt starken Einflüssen des Mediums entlang des Wellenausbreitungsweges in vulkanischen Gebieten liegen. <br /> Insgesamt ist die Erkennungsleistung des entwickelten automatischen Klassifizierungssystems als sehr vielversprechend einzustufen. Im Gegensatz zu Standardverfahren, bei denen in der Seismologie üblicherweise nur der Startzeitpunkt eines seismischen Ereignisses detektiert wird, werden in dem untersuchten Verfahren seismische Ereignisse in ihrer Gesamtheit erfasst und zudem im selben Schritt bereits klassifiziert. / Merapi volcano is one of the most active and dangerous volcanoes of the earth. Located in central part of Java island (Indonesia), even a moderate eruption of Merapi poses a high risk to the highly populated area. Due to the close relationship between the volcanic unrest and the occurrence of seismic events at Mt. Merapi, the monitoring of Merapi's seismicity plays an important role for recognizing major changes in the volcanic activity. An automatic seismic event detection and classification system, which is capable to characterize the actual seismic activity in near real-time, is an important tool which allows the scientists in charge to take immediate decisions during a volcanic crisis. <br /> In order to accomplish the task of detecting and classifying volcano-seismic signals automatically in the continuous data streams, a pattern recognition approach has been used. It is based on the method of hidden Markov models (HMM), a technique, which has proven to provide high recognition rates at high confidence levels in classification tasks of similar complexity (e.g. speech recognition). Any pattern recognition system relies on the appropriate representation of the input data in order to allow a reasonable class-decision by means of a mathematical test function. Based on the experiences from seismological observatory practice, a parametrization scheme of the seismic waveform data is derived using robust seismological analysis techniques. The wavefield parameters are summarized into a real-valued feature vector per time step. The time series of this feature vector build the basis for the HMM-based classification system. In order to make use of discrete hidden Markov (DHMM) techniques, the feature vectors are further processed by applying a de-correlating and prewhitening transformation and additional vector quantization. The seismic wavefield is finally represented as a discrete symbol sequence with a finite alphabet. This sequence is subject to a maximum likelihood test against the discrete hidden Markov models, learned from a representative set of training sequences for each seismic event type of interest.<br /> A time period from July, 1st to July, 5th, 1998 of rapidly increasing seismic activity prior to the eruptive cycle between July, 10th and July, 19th, 1998 at Merapi volcano is selected for evaluating the performance of this classification approach. Three distinct types of seismic events according to the established classification scheme of the Volcanological Survey of Indonesia (VSI) have been observed during this time period. Shallow volcano-tectonic events VTB (h < 2.5 km), very shallow dome-growth related seismic events MP (h < 1 km) and seismic signals connected to rockfall activity originating from the active lava dome, termed Guguran.<br /> The special configuration of the digital seismic station network at Merapi volcano, a combination of small-aperture array deployments surrounding Merapi's summit region, allows the use of array methods to parametrize the continuously recorded seismic wavefield. The individual signal parameters are analyzed to determine their relevance for the discrimination of seismic event classes. For each of the three observed event types a set of DHMMs has been trained using a selected set of seismic events with varying signal to noise ratios and signal durations. Additionally, two sets of discrete hidden Markov models have been derived for the seismic noise, incorporating the fact, that the wavefield properties of the ambient vibrations differ considerably during working hours and night time. <br /> A total recognition accuracy of 67% is obtained. The mean false alarm (FA) rate can be given by 41 FA/class/day. However, variations in the recognition capabilities for the individual seismic event classes are significant. Shallow volcano-tectonic signals (VTB) show very distinct wavefield properties and (at least in the selected time period) a stable time pattern of wavefield attributes. The DHMM-based classification performs therefore best for VTB-type events, with almost 89% recognition accuracy and 2 FA/day. <br /> Seismic signals of the MP- and Guguran-classes are more difficult to detect and classify. Around 64% of MP-events and 74% of Guguran signals are recognized correctly. The average false alarm rate for MP-events is 87 FA/day, whereas for Guguran signals 33 FA/day are obtained. However, the majority of missed events and false alarms for both MP and Guguran events are due to confusion errors between these two event classes in the recognition process. <br /> The confusion of MP and Guguran events is interpreted as being a consequence of the selected parametrization approach for the continuous seismic data streams. The observed patterns of the analyzed wavefield attributes for MP and Guguran events show a significant amount of similarity, thus providing not sufficient discriminative information for the numerical classification. The similarity of wavefield parameters obtained for seismic events of MP and Guguran type reflect the commonly observed dominance of path effects on the seismic wave propagation in volcanic environments.<br /> The recognition rates obtained for the five-day period of increasing seismicity show, that the presented DHMM-based automatic classification system is a promising approach for the difficult task of classifying volcano-seismic signals. Compared to standard signal detection algorithms, the most significant advantage of the discussed technique is, that the entire seismogram is detected and classified in a single step.

In mobile speech communication, adverse conditions, such as noisy acoustic environments and unreliable network connections, may severely degrade the intelligibility and natural- ness of the received speech quality, and increase the listening effort. This thesis focuses on countermeasures based on statistical signal processing techniques. The main body of the thesis consists of three research articles, targeting two specific problems: speech enhancement for noise reduction and flexible source coder design for unreliable networks. Papers A and B consider speech enhancement for noise reduction. New schemes based on an extension to the auto-regressive (AR) hidden Markov model (HMM) for speech and noise are proposed. Stochastic models for speech and noise gains (excitation variance from an AR model) are integrated into the HMM framework in order to improve the modeling of energy variation. The extended model is referred to as a stochastic-gain hidden Markov model (SG-HMM). The speech gain describes the energy variations of the speech phones, typically due to differences in pronunciation and/or different vocalizations of individual speakers. The noise gain improves the tracking of the time-varying energy of non-stationary noise, e.g., due to movement of the noise source. In Paper A, it is assumed that prior knowledge on the noise environment is available, so that a pre-trained noise model is used. In Paper B, the noise model is adaptive and the model parameters are estimated on-line from the noisy observations using a recursive estimation algorithm. Based on the speech and noise models, a novel Bayesian estimator of the clean speech is developed in Paper A, and an estimator of the noise power spectral density (PSD) in Paper B. It is demonstrated that the proposed schemes achieve more accurate models of speech and noise than traditional techniques, and as part of a speech enhancement system provide improved speech quality, particularly for non-stationary noise sources. In Paper C, a flexible entropy-constrained vector quantization scheme based on Gaus- sian mixture model (GMM), lattice quantization, and arithmetic coding is proposed. The method allows for changing the average rate in real-time, and facilitates adaptation to the currently available bandwidth of the network. A practical solution to the classical issue of indexing and entropy-coding the quantized code vectors is given. The proposed scheme has a computational complexity that is independent of rate, and quadratic with respect to vector dimension. Hence, the scheme can be applied to the quantization of source vectors in a high dimensional space. The theoretical performance of the scheme is analyzed under a high-rate assumption. It is shown that, at high rate, the scheme approaches the theoretically optimal performance, if the mixture components are located far apart. The practical performance of the scheme is confirmed through simulations on both synthetic and speech-derived source vectors. / QC 20100825

A Mandarin speech recognition system for addresses based on MFCC, hidden Markov model (HMM) and Viterbi algorithm is proposed in this thesis. HMM is a doubly stochastic process describing the ways of pronunciation by recording the state transitions according to the time-varing properties of the speech signal. In order to simplify the system design and reduce the computational cost, the mono-syllable structure information in Mandarin is used by incorporating both mono-syllable recognizor and HMM for our system. For the speaker-dependent case, Mandarin address inputting can be accomplished within 60 seconds and 98% correct identification rate can be achieved in the laboratory environment.

A Mandarin speech recognition system for addresses in Taiwan, based on end-point detection, MFCC and HMM, is proposed and implemented in this thesis. It includes both phrase and monosyllable recognition tasks. For the phrase recognition part, we select the initial candidates before the final recognition stage to tremendously reduce the computational time. On the other side, for the monosyllable recognition part, we further refine the recognition details to improve the correct rate under easily confused circumstances. The final system can achieve 85% correct identification rate, and the address recognition can be completed within 2 seconds in the laboratory environment for speaker-dependent case.

A system of Chinese resume by speech construction is developed by the use of a novel segmentation mechanism and the classical Hidden Markov Model. The recognition system is based on both mono-syllable HMM's and speech-text alignment schemes. Experimental results indicate that the amount of training materials used for feature extraction can be greatly reduced, and the text content of the recorded speech training data can be different from those of the recognition tasks as well. Each phrase in the resume can be identified within one second, that is approximately the same as the graduate did last year. Furthermore, the user interface of the resume system has been redesigned and polished by the GTK toolkit in order to enable event-driven X-window operations.

A design of speech recognition system for Chinese names of historical figures around the world is proposed in this thesis. A speech database of approximately forty-six thousand Chinese names is collected and recorded twice for system evaluation. This system applies Mel-frequency cepstrum coefficients, monosyllable HMM¡¦s and speech-text alignment scheme to accomplish initial candidate selection. A Mandarin pitch identification mechanism is then followed to increase the correct rate and obtain the final answer. The experimental results indicate that a 90% correct identification rate can be achieved, under the condition that the first session recording material is used for training and the second one for testing. For the speaker dependent case, the correct name can be recognized within 1.5 seconds, using a PC with an Intel Celeron 2.4 GHz CPU and RedHat Linux 9.0 Operation System.

No description available.

Speech is intuitive, fast and easy to generate, but it is hard to index and easy to forget. What is more, listening to speech is slow. Text is easier to store, process and consume, both for computers and for humans, but writing text is slow and requires some intention. In this thesis, we study speech recognition which allows converting speech into text, making it easier both to create and to use information. Our tool of study is Hidden Markov Models which is one of the most important machine learning models in speech and language processing. The aim of this thesis is to do a rst study in Hidden Markov Models and understand their importance, particularly in speech recognition. We will go through three fundamental problems that come up naturally with Hidden Markov Models: to compute a likelihood of an observation sequence, to nd an optimal state sequence given an observation sequence and the model, and to adjust the model parameters. A solution to each problem will be given together with an example and the corresponding simulations using MatLab. The main importance lies in the last example, in which a rst approach to speech recognition will be done.

<p>The decision making processes in today's manufacturing systems represent very complex and challenging tasks. The desired flexibility in terms of the functionality of a machine adds more components to the machine. The real time monitoring and reporting generates large streams of data. However the intelligent and real time processing of this large collection of system data is at the core of the manufacturing decision support tools. </p> <p>This thesis outlines the use of Frequent Episodes in Event Sequences and Hidden Markov Modeling of throughput, quality and maintenance data to model the deterioration of performance in the components that make up the manufacturing system. The thesis also introduces the concept of decision points and outlines how to integrate the total cost function in a business model. </p> This thesis deals with the following three topics: <p>First, the component-based data structure of the manufacturing system is outlined especially throughput, quality and maintenance data. In this approach, the manufacturing system is considered as a group of components that interact with each other and with raw materials to produce the manufactured product. This interaction creates a considerable amount of data which can be associated with the relevant components of the system. The relations between the manufacturing components are established on a physical and logical basis. The components properties are clearly defined in database tables specifically created for this application. The thesis also discusses the web services in manufacturing systems and the portable technologies used in plant decision support tools. </p> <p>Second, the thesis presents a novel application of Frequent Episodes in Event Sequences to identify patterns in the deterioration of performance in a component using frequent episodes of operational failures, quality failures and maintenance activities. A Hidden Markov Model (HMM) is used to model each deterioration episode to estimate the states of performance and the transition rates between the states. The thesis compares the results generated by this model to other existing models of component performance deterioration while emphasizing the benefits ofthe proposed model through the use of the plant data.</p> <p>Finally the thesis presents a methodology usmg HMM probability distributions and Bayesian Decision theory framework to provide a set of decisions and recommendations under the condition of data uncertainty. The results of this analysis are then integrated in the plant maintenance business model.</p> <p>It is worthwhile mentioning that to develop the techniques and validate the results in this research; a Manufacturing Execution System (MES) was developed to operate in an automotive engine plant. All the data and results in this research are based on the plant data. The MES which was developed in this research provided significant benefits in the plant and was adapted by many other GM plants around the world.</p> / Thesis / Doctor of Philosophy (PhD)