In the realm of computer science and information technology, scientific methodologies can be categorized into two overarching paradigms: analytical approaches and strategies grounded in machine learning or optimization techniques. In the context of machine learning, the massive employment of neural&#x...

This dissertation focuses on new architectures for computational speech generation using neural networks. With their advent, there has been a very vigorous development of new methods that have enabled the generation of speech with higher quality and naturalness than traditional methods have&...

The work was focused on the research and development of methods, software and the necessary components of the measuring chain for the purpose of analysis and subsequent monitoring of the torsional interaction of the turbine-generator with the events in the electricity network. As&...

The quantification of the dynamic force coefficients of labyrinth seals used in steam turbines is currently not clearly resolved. Numerical tools offer the calculation of seal coefficients in the design of the turbomachinery; however, a lack of experimental data makes comparison and&#x...

This dissertation deals in detail with diagnostics and monitoring of steam turbine rotating blades. Specifically, it focuses on the rotating blades of the last bladed wheel of the low-pressure turbine. Blades installed in this part are forced to vibrate at most due to their si...

Seam tracking systems are used for automatic guidance of welding robots. They have been developed for several decades and offer higher accuracy, robustness and time efficiency compared to the manual welding process. In addition, the knowledge acquired in this area is further transformed...

The thesis focuses on the topic of speaker diarization, a speech processing task that is commonly characterized as the question "Who speaks when?". It also addresses the related task of overlapping speech detection, which is very relevant for diarization. The theoretic...

This work contributes with novel computer vision methods and datasets for the automatic identification and localization of biological specimens in a natural environment (i.e. "in the wild"). With great emphasis placed on suitability and application in biodiversity monitoring, conservation, a...

This dissertation thesis deals with the problem of environment changes in the tasks of mobile robotics. In particular, it focuses on using of one-dimensional non-visual sensors in order to reduce computation cost. The work presents a new system for detection and classification of t...

This thesis has five main goals. At first, it maps the datasets used for long-term visual localization and selects viable datasets for further evaluation. Next, one of the current state-of-the-art pipelines is selected and enhanced. Results with carefully fine-tuned methods' parameters a...

The predictive control is currently one of the most popular control strategies. The main benefits of the predictive control include the ability to satisfy the constraints with less effort than in case of other strategies. On the other hand, the predictive control hardly manages...

The thesis deals with the estimation of noise properties of the state space model. The thesis presents four basic approaches to the estimation of~the~noise properties, where the special attention is devoted to the estimation using correlation methods. Also, the detailed ways of calculat...

This dissertation thesis deals with speech recognition of patients who have undergone total laryngectomy and who use electrolarynx for voice production. In the first part of the thesis, the reasons of voice loss and methods that are currently used for its rehabilitation together wi...

This dissertation thesis presents a novel system for automatic heterogeneous face recognition from facial sketches based on a novel method named X-Bridge. Such a task is primarily relevant in the security and surveillance domains. In this work are made following contributions: (1) Analy...

The aim of this is thesis is the creation of new visual features for the automatic lipreading systems. State-of-the-art methods (mainly machine learning methods) are not using any form of adaptation for a specific speaker during their training. We ask, how to adapt this method...

Gaussian process regression is a Bayesian nonparametric model, which combines high expressiveness with tractable Bayesian inference. The first application of GP models in this thesis is the identification of nonlinear time-invariant systems affine in control with functional uncertainty.&...

Blade Tip-Timing measurement becomes more and more valuable because power generation market is changing to the peak-load market. Today steam turbines have to full-fill high efficiency and reliability demand as well as increasing of number of cycles or operating in wide operational range...

The research area of this thesis is online topic-based language model (LM) adaptation. It is a mechanism designed to reduce word error rates of real-time automatic speech recognition (ASR) systems in multi-topic tasks, where a general LM cannot model varying word sequence statistics...

This work focuses on development of innovative interval based methods in the area of power system optimization with respect to changing environmental, technological, and market conditions. In the review part, a thorough introduction into the history of power network computation was given.&#x...

The main part of the dissertation is devoted to a development of the active fault detection on the infinite time horizon. Simultaneously, the thesis deals briefly with data-driven passive fault detection methods that can be used when the system model is not known in advance.&#...