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.

Hyperspectral signature classification is a kind of quantitative analysis approach for hyperspectral imagery which performs detection and classification of the constituent materials at pixel level in the scene. The classification procedure can be operated directly on hyperspectral data or performed by using some features extracted from corresponding hyperspectral signatures containing information like signature energy or shape. In this paper, we describe a technique that applies non-homogeneous hidden Markov chain (NHMC) models to hyperspectral signature classification. The basic idea is to use statistical models (NHMC models) to characterize wavelet coefficients which capture the spectrum structural information at multiple levels. Experimental results show that the approach based on NHMC models outperforms existing approaches relevant in classification tasks.

The process of locating moving objects through video sequences is a fundamental computer vision problem. This process is referred to as video tracking and has a broad range of applications. Even though video tracking is an open research topic that have received much attention during recent years, developing accurate and robust algorithms that can handle complicated tracking tasks and scenes is still challenging. One challenge in computer vision is to develop systems that like humans can understand, interpret and recognize visual information in different situations. In this master thesis work, a tracking algorithm based on eye tracking data is proposed. The aim was to compare the tracking performance of the proposed algorithm with a state-of-the-art video tracker. The algorithm was tested on gaze signals from five participants recorded with an eye tracker while the participants were exposed to dynamic stimuli. The stimuli were moving objects displayed on a stationary computer screen. The proposed algorithm is working offline meaning that all data is collected before analysis. The results show that the overall performance of the proposed eye tracking algorithm is comparable to the performance of a state-of-the-art video tracker. The main weaknesses are low accuracy for the proposed eye tracking algorithm and handling of occlusion for the video tracker. We also suggest a method for using eye tracking as a complement to object tracking methods. The results show that the eye tracker can be used in some situations to improve the tracking result of the video tracker. The proposed algorithm can be used to help the video tracker to redetect objects that have been occluded or for some other reason are not detected correctly. However, ATOM brings higher accuracy.

In the last decades research and development in the field of robotics has grown rapidly. This growth has resulted in the emergence of service robots that need to be able to physically interact with humans for different applications. One of these applications involves robots and humans cooperating in handling an object together. In such cases, there is usually an initial arrangement of how the robot and the humans hold the object and the arrangement stays the same throughout the manipulation task. Real-world scenarios often require that the initial arrangement changes throughout the task, therefore, it is important that the robot is able to recognize these changes and act accordingly. We consider a setting where a robot holds a large flat object with one or two humans. The aim of this research project is to detect the change in the number of agents grasping the object using only force and torque information measured at the robot's wrist. The proposed solution involves defining a transition sequence of four steps that the humans should perform to go from the initial scenario to the final one. The force and torque information is used to estimate the grasping point of the agents with a Kalman filter. While the humans are going from one scenario to the other, the estimated point changes according to the step of the transition the humans are in. These changes are used to track the steps in the sequence using a hidden Markov model (HMM). Tracking the steps in the sequence means knowing how many agents are grasping the object. To evaluate the method, humans that were not involved in the training of the HMM were asked to perform two tasks: a) perform the previously defined sequence as is, and b) perform a deviation of the sequence. The results of the method show that it is possible to detect the change between one human and two humans holding the object using only force and torque information.

京都大学 / 新制・課程博士 / 博士(情報学) / 甲第23321号 / 情博第757号 / 新制||情||129(附属図書館) / 京都大学大学院情報学研究科数理工学専攻 / (主査)教授 太田 快人, 教授 山下 信雄, 教授 大塚 敏之 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DGAM

La quantité de données audio disponibles, telles que les enregistrements radio, la musique, les podcasts et les publicités est en augmentation constance. Par contre, il n'y a pas beaucoup d'outils de classification et d'indexation, qui permettent aux utilisateurs de naviguer et retrouver des documents audio. Dans ces systèmes, les données audio sont traitées différemment en fonction des applications. La diversité de ces techniques d'indexation rend inadéquat le traitement simultané de flux audio où différents types de contenu audio coexistent. Dans cette thèse, nous présentons nos travaux sur l'extension de l'approche ALISP, développé initialement pour la parole, comme une méthode générique pour l'indexation et l'identification audio. La particularité des outils ALISP est qu'aucune transcription textuelle ou annotation manuelle est nécessaire lors de l'étape d'apprentissage. Le principe de cet outil est de transformer les données audio en une séquence de symboles. Ces symboles peuvent être utilisés à des fins d'indexation. La principale contribution de cette thèse est l'exploitation de l'approche ALISP comme une méthode générique pour l'indexation audio. Ce système est composé de trois modules: acquisition et modélisation des unités ALISP d'une manière non supervisée, transcription ALISP des données audio et comparaison des symboles ALISP avec la technique BLAST et la distance de Levenshtein. Les évaluations du système proposé pour les différentes applications sont effectuées avec la base de données YACAST et avec d'autres corpus disponibles publiquement pour différentes tâche de l'indexation audio. / The amount of available audio data, such as broadcast news archives, radio recordings, music and songs collections, podcasts or various internet media is constantly increasing. Therefore many audio indexing techniques are proposed in order to help users to browse audio documents. Nevertheless, these methods are developed for a specific audio content which makes them unsuitable to simultaneously treat audio streams where different types of audio document coexist. In this thesis we report our recent efforts in extending the ALISP approach developed for speech as a generic method for audio indexing, retrieval and recognition. The particularity of ALISP tools is that no textual transcriptions are needed during the learning step. Any input speech data is transformed into a sequence of arbitrary symbols. These symbols can be used for indexing purposes. The main contribution of this thesis is the exploitation of the ALISP approach as a generic method for audio indexing. The proposed system consists of three steps; an unsupervised training to model and acquire the ALISP HMM models, ALISP segmentation of audio data using the ALISP HMM models and a comparison of ALISP symbols using the BLAST algorithm and Levenshtein distance. The evaluations of the proposed systems are done on the YACAST and other publicly available corpora for several tasks of audio indexing.

Robotics in rehabilitation provides considerable opportunities to improve the quality of life for persons with disabilities. Computerized and Virtual Environment (VE) training systems for persons with disabilities, many of which utilize the haptic feedback, have gained increasing acceptance in the recent years. Our methodology here is based on creating virtual environments connected to a haptic interface as an input device. This robotic setup introduces the advantages of the haptic rendering features in the environment and also provides tactile feedback to the patients. This thesis aims to demonstrate the efficacy of assistance function algorithms in rehabilitation robotics in virtual environments. Assist functions are used to map limited human input to motions required to perform complex tasks. The purpose is to train individuals in task-oriented applications to insure that they can be incorporated into the workplace. Further, Hidden Markov Model (HMM) based motion recognition and skill learning are used for improving the skill levels of the users. For the Hidden Markov Model based motion recognition, the user's motion intention is combined with environment information to apply an appropriate assistance function. We used this algorithm to perform a commonly used vocational therapy test referred to as the box and the blocks test. The Hidden Markov Model based skill approach can be used for learning human skill and transferring the skill to persons with disabilities. A relatively complex task of moving along a labyrinth is chosen as the task to be modeled by HMM. This kind of training allows a person with disability to learn the skill and improve it through practice. Its application to motion therapy system using a haptic interface helps in improving their motion control capabilities, tremor reduction and upper limb coordination. The results obtained from all the tests demonstrated that various forms of assistance provided reduced the execution times and increased the motion performance in chosen tasks. Two persons with disabilities volunteered to perform the above tasks and both of the disabled subjects expressed an interest and satisfaction with the philosophy behind these concepts.

Indiana University-Purdue University Indianapolis (IUPUI) / It is generally of interest to explore if the risk of death would be modified by medical conditions (e.g., illness) that have occurred prior. This situation gives rise to semicompeting risks data, which are a mixture of competing risks and progressive state data. This type of data occurs when a non-terminal event can be censored by a well-defined terminal event, but not vice versa. In the first part of this dissertation, the shared gamma-frailty conditional Markov model (GFCMM) is adopted because it bridges the copula models and illness-death models. Maximum likelihood estimation methodology has been proposed in the literature. However, we found through numerical experiments that the unrestricted model sometimes yields nonparametric biased estimation. Hence a practical guideline is provided for using the GFCMM that includes (i) a score test to assess whether the restricted model, which does not exhibit estimation problems, is reasonable under a proportional hazards assumption, and (ii) a graphical illustration to evaluate whether the unrestricted model yields nonparametric estimation with substantial bias for cases where the test provides a statistical significant result against the restricted model. However, the scientific question of interest that whether the status of non-terminal event alters the risk to terminal event can only be partially addressed based on the aforementioned approach. Therefore in the second part of this dissertation, we adopt a Markov illness-death model, whose transition intensities are essentially equivalent to the marginal hazards defined in GFCMM, but with different interpretations; we develop three nonparametric tests, including a linear test, a Kolmogorov-Smirnov-type test, and a L2-distance-type test, to directly compare the two transition intensities under consideration. The asymptotic properties of the proposed test statistics are established using empirical process theory. The performance of these tests in nite samples is numerically evaluated through extensive simulation studies. All three tests provide similar power levels with non-crossing curves of cumulative transition intensities, while the linear test is suboptimal when the curves cross. Eventually, the proposed tests successfully address the scientific question of interest. This research is applied to Indianapolis-Ibadan Dementia Project (IIDP) to explore whether dementia occurrence changes mortality risk. / 2022-05-06

An early detection of heart disease can save lives, caution individuals and also help to determine the type of treatment to be given to the patients. The first test of diagnosing a heart disease is through auscultation - listening to the heart sounds. The interpretation of heart sounds is subjective and requires a professional skill to identify the abnormalities in these sounds. A medical practitioner uses a stethoscope to perform an initial screening by listening for irregular sounds from the patient's chest. Later, echocardiography and electrocardiography tests are taken for further diagnosis. However, these tests are expensive and require specialized technicians to operate. A simple and economical way is vital for monitoring in homecare or rural hospitals and urban clinics. This dissertation is focused on developing a patient-centered device for initial screening of the heart sounds that is both low cost and can be used by the users on themselves, and later share the readings with the healthcare providers. An innovative mobile health service platform is created for analyzing and classifying heart sounds. Certain properties of heart sounds have to be evaluated to identify the irregularities such as the number of heart beats and gallops, intensity, frequency, and duration. Since heart sounds are generated in low frequencies, human ears tend to miss certain sounds as the high frequency sounds mask the lower ones. Therefore, this dissertation provides a solution to process the heart sounds using several signal processing techniques, identifies the features in the heart sounds and finally classifies them. This dissertation enables remote patient monitoring through the integration of advanced wireless communications and a customized low-cost stethoscope. It also permits remote management of patients' cardiac status while maximizing patient mobility. The smartphone application facilities recording, processing, visualizing, listening, and classifying heart sounds. The application also generates an electronic medical record, which is encrypted using the efficient elliptic curve cryptography and sent to the cloud, facilitating access to physicians for further analysis. Thus, this dissertation results in a patient-centered device that is essential for initial screening of the heart sounds, and could be shared for further diagnosis with the medical care practitioners.

No description available.