National Repository of Grey Literature
Not a member yet
574846 research outputs found
Sort by
Far-Field Speaker Verification Incorporating Multichannel Processing
No full textZpracování řeči ze vzdálených zdrojů upoutává v posledních letech stále větší pozornost díky vzestupu inteligentních reproduktorů, domácích asistentů nebo systémů pro přepisování schůzek. K podpoře těchto aplikací jsou zapotřebí robustní techniky pro zpracování řeči ze vzdálených zdrojů. Klíčem k personalizované interakci je verifikace mluvčího. Ve srovnání se zpracováním signálů získaných mikrofony umístěnými blízko zdrojů čelí systémy zpracovávající nahrávky ze vzdálených senzorů dalším výzvám. Jimi jsou jevy, které zhoršují kvalitu cílové řeči, zejména dozvuk a šum v pozadí. Zařízení pro vzdálené snímání proto často využívají mikrofonní pole, která mohou napomoci zmírnění uvedených jevů díky tomu, že poskytují informaci o prostoru. Nastíněné výzvy a příležitosti motivují tuto práci, jež se zaměřuje na vícekanálovou verifikaci mluvčího. Z důvodu omezených zdrojů dat zůstává vícekanálová verifikace mluvčích nedostatečně prozkoumaná, a to navzdory významnému pokroku v souvisejících oblastech zpracování řeči. Předložená práce se věnuje dvěma fundamentálním aspektům: absenci dat i specializovaným technikám zpracování. V rámci problematiky dat jsme přetvořili existující veřejně dostupné datové sady a vyvinuli novou sadu, MultiSV, která poskytuje simulované vícekanálové směsi signálů s trénovacími referenčními signály řeči/šumu a identitami mluvčích. MultiSV rovněž obsahuje znovu přenesené nahrávky využitelné pro vyhodnocování verifikačních systémů s podporou různých scénářů. Příkladem může být varianta s registrací do systému pomocí jednokanálových čistých nebo vícekanálových poškozených nahrávek. Abychom podpořili trénování modelů s vyššími nároky na data, vytvořili jsme rozšířenou datovou sadu MultiSV2. Na úrovni modelů jsme nejprve přistoupili k extrakci řečových embeddingů z vícekanálového audia pomocí kaskádové strategie, přičemž jsme problém rozložili na vícekanálové předzpracování a jednokanálovou extrakci embeddingů. Motivováni pokroky v oblasti separace řeči jsme navrhli různé techniky od těch založených na zpracování signálů po hybridní, které kombinují neuronové sítě a beamforming. V práci se diskutuje přímá a nepřímá predikce masek pro beamforming na nich založený. Dále je představena metoda RCA (reference channel attention), která zobecňuje jednokanálové separační modely tak, aby mohly využít vícekanálové vstupy. S uvědoměním si omezení kaskádových modelů, jimiž jsou šíření chyb a odlišné objektivní funkce jednotlivých modulů, jsme dále prozkoumali jednotné architektury pro vícekanálovou extrakci embeddingů. S využitím MultiSV2 jsme dokázali ladit parametry spojených komponent kaskádových modelů s využitím cílové objektivní funkce. Následně jsme navrhli techniku METRO (multi-channel extension of pre-trained models), která rozšiřuje původně jednokanálové modely pro reprezentaci řeči trénované pomocí samořídícího učení na vícekanálové. Ačkoli METRO v kombinaci s vhodným extrakčním modelem produkuje řečové embeddingy pro vícekanálové audio, metoda samotná je obecná a potenciálně využitelná v dalších oblastech zpracování řeči
Effect of postprocessing on structure and properties alternatively prepared metallic materials
No full textAlternative manufacturing processes, such as additive technologies, are currently the subject of intensive research because they represent innovative approaches with potential advantages compared to conventional manufacturing methods. On the other hand, for their broader commercial application, it is necessary to address certain issues accompanying these processes. Among the best-known additive technologies is 3D printing, but materials produced in this way may exhibit a certain degree of porosity, unfavourable residual stress states and microstructures, and consequently lower static or dynamic properties, whether at room or elevated temperatures. One possibility for optimizing these challenges is post-processing of such materials. In this work, the research focuses on methods such as thermal or thermo-mechanical treatment and their effect on selected properties of metallic materials produced by alternative processes (SLM technology). The investigated materials comprised the tool steel Dievar, AISI H13, and the nickel-based superalloy Inconel 718. The results indicate that thermo-mechanical processing has a beneficial effect on microstructure evolution, modification of residual stress within the material, reduction of porosity, and enhancement of mechanical properties. To evaluate the effect of post-processing, conventional as well as non-conventional methods were employed, such as accelerated creep testing
Methods for Analysis of Long-Term Clinical Recordings of Intracranial EEG in Epileptic Patients
No full textIntracranial EEG (iEEG) is an indispensable method for pre-surgical planning in patients with pharmacoresistant epilepsy. The current clinical standard for the identification of surgical targets is grounded in the analysis of ictal data, while interictal recordings still play a minor role in clinical diagnoses despite representing 99% of presurgical data. This thesis introduces relative entropy as a new feature for the analysis of interictal data to locate the epileptogenic zone, hence the surgical target. It provides analyses of confounding variables such as antiseizure medication, states of vigilance, and proximity to seizures and reveals that states of vigilance modulate feature values and distributions the most. It further investigates novel features of very-fast ripples and ultra-fast ripples and explores their behavior in different states of vigilance. Finally, a novel detection method for ultra-fast oscillations (2-8kHz) was designed and implemented to allow quantitave analyses of this novel phenomena. Together, these results offer new features for the identification of the epileptogenic zone and contribute to a deeper understanding of presurgical iEEG recordings by description of different confounding factors and their influence on iEEG
Development od Modified Binders and Asphalt Mixtures with Optimized Grain Size Distribution
No full textDissertation focuses on the comprehensive evaluation of asphalt binders and coarse asphalt mixtures. The first part of the study investigates the synergistic effects of combining polyphosphoric acid (PPA) and styrene-butadiene-styrene (SBS) as modifier additives in asphalt binders, aiming to reduce the overall content of used modification and consequently lower emissions of the modified binder. The study examines whether the combination of modifiers can enhance binder stability at high temperatures, improve low-temperature properties, and increase resistance to aging. The research revealed that the modifications exhibit synergistic effects in some cases, providing performance benefits compared to their individual use. The second part of the dissertation focuses on the development and evaluation of the coarse-grained mixture ACP 32, with the aim of utilizing coarse aggregates and relieving industrial pressure on scarce aggregate fractions. The mixture variants were tested for stiffness modulus, fatigue characteristics, resistance to permanent deformation, and low-temperature resistance. The results demonstrated that the ACP 32 mixture can meet the requirements of Czech technical regulations for mixture ACP 22, enabling its classification among mixtures for base layers
Intelligent Tool for Content Analysis of Specific Audiovisual Production
No full textThis doctoral thesis focuses on developing an intelligent tool for analysing the content of specific audiovisual production by means of modern artificial intelligence techniques. It answers the growing demand for high-quality automated analysis of artistic audiovisual pieces whose distinctive features render conventional analytic procedures inadequate and time-consuming. The primary objective is to design a robust classification methodology for both visual and audio content and to implement an interactive web-based application that allows art historians to examine bibliographic and content data efficiently. The study also introduces a thoroughly annotated data set with specially defined categories for training and validating advanced CNN models. The experimental section evaluates the performance of a pre-trained CNN together with VGG, Inception, Xception and NASNet architectures fine-tuned on the curated data set. For audio analysis, CNN 14, ResNet and Wavegram networks are employed. The results verify the suitability of the chosen methods for the data under investigation. Finally, the work delivers a comprehensive, user-friendly tool that supports interactive exploration of audiovisual material and provides a dedicated interface for detailed data set construction
Reflection of Socialism in Fictive and Factual Narativ of Pavel Tigrid
No full textThe thesis focuses on the analysis of fictive and factual narrative in Pavel Tigrid's texts thematically focused on the reflection of socialism in Czechoslovakia. Methodologically, the thesis will be based mainly on Hayden White's texts oriented to the given issue and on Ralf Dahrendorf's theories focused on the typology of the engaged intellectual. The methodology will be complemented by Seymour Chatman's model focused on the components of narrative theory, according to whose existents analyses of the author's texts will be developed, namely the novel Pocket Guide of an Intelligent Woman to Her Own Destiny and the essay Marx at Hradčany. The ambition of the thesis will be to find the specifics of fictional and non-fictional narrative in these selected texts by Pavel Tigrid and to point out to what extent the two types of narratives in specific texts resemble each other, overlap, or are related
The political and cultural activity of the city of Volyně between 1918 and 1948
No full textThis thesis focuses on the development of the town of Volyně during the period from 1918 to 1948, while also including, for broader context, the history and development of the town from its founding up to the beginning of the studied period in 1918. The work also examines individual contemporary records and their changes based on the periods in which they were created. In addition to written sources, many period photographs as well as aerial and cadastral maps are utilized. The introduction summarizes the topic of the thesis, including an overview of the sources and literature used. The first chapter deals with the history of the town prior to 1918. The second chapter gradually analyzes the development of the town between 1918 and 1948. Due to its extent, this chapter is divided into separate subchapters focusing on specific periods within these years: the establishment of Czechoslovakia and the so-called First Republic 1918-1938, the Second Republic and the Protectorate period (1938-1945), and the post-war period known as the Third Republic 1945-1948. The chapter also contains a separate subchapter dedicated to the architectural and demographic development of the town based on available data, covering the years 1837-1948. The remainder of the thesis then focuses on the analysis of contemporary sources and the comparison of significant events mentioned in local chronicles, examining how these records reflected the changes of the time. It thus concentrates on the differences and similarities between individual entries and to what extent the records of various authors differed from one another. For better clarity, this chapter is also divided into separate subchapters devoted to selected events
Measurement and analysis of light smog in selected locations
No full textThis bachelor's thesis focuses on identifying locations suitable for astronomical observations. For their selection, maps displaying light pollution from the year 2015 were used, specifically the Light Pollution Map. Subsequently, more specific areas were selected with regard to terrain conditions, vegetation, and ccessibility. The identification of particular sites was carried out using aerial photographs and tourist maps. The final locations were chosen based on personal photographic documentation from each site and written notes recorded on-site. Light pollution was measured using the SQM photometer by the Canadian company Unihedron, which captures the night sky within a 120° field of view. The result of the measurement is a value indicating the sky brightness in units of mag/arcsec?. The higher the measured value, the darker the sky. Measurements were conducted at seven locations, which were identified in the districts of Tábor, Písek, Příbram, Strakonice, and Benešov. Specifically, these are Komárov, Žďákov Bridge, Nadějkov, Podskalí, Hutě pod Třemšínem, Nevželice, and Kamberk. The highest values were recorded at the Žďákov Bridge, where a value of 21.31 mag/arcsec? was measured. The same result, 21.31 mag/arcsec?, was also measured at Podskalí. The third highest value was measured near the village of Kamberk, specifically 21.29 mag/arcsec?
Development of dramatic activities in preschool children by folk songs
No full textThe bachelor thesis focuses on the connection between drama and musical activities in preschool children through folk songs. The aim is to highlight the possibilities of using these songs as a motivational and developmental tool in dramatic play. The theoretical part focuses on the developmental specifics of preschool children, drama education, and musical activities within early childhood education. It also discusses the characteristics of folk songs and their potential in working with children. The practical part includes the design and implementation of four thematic units, whose activities are based on specific folk songs and combine elements of drama and music. During the implementation, observation and interview methods are used to track children's sincere reactions, engagement, emotional experiences, and expressions. The results show that drama activities based on folk songs promote creativity, communication, cooperation and pro-social behaviour. The thesis recommends more frequent and varied use of songs during the thematic unit and confirms the benefits of linking music and drama in preschool education
A musical fairy tale as a means of developing musical abilities and skills with the help of dramatic education methods.
No full textThis thesis focuses on the development of musical abilities and skills in preschool children through the use of drama education methods. The aim of the thesis is to verify the possibility of developing musical abilities using the musical fairy tale The Lost Handkerchief. The thesis is divided into two parts. The theoretical part defines the concepts of musical fairy tale, musical activities of preschool children, songs, and drama education. The practical part includes a program of musical activities, where singing, instrumental, and dramatic activities contribute to the creation of a musical fairy tale using folk songs. This program is implemented in a kindergarten setting