Pomocí matematického formalismu váhovaných konečných stavových převodníků (weighted finite state transducers WFST) může být formulována řada úloh včetně automatického rozpoznávání řeči (automatic speech recognition ASR). Dnešní ASR systémy široce využívají složených pravděpodobnostních modelů nazývaných dekódovací grafy nebo rozpoznávací sítě. Ty jsou z jednotlivých komponent konstruovány pomocí WFST operací, např. kompozice. Každá komponenta je zde zdrojem znalostí a omezuje vyhledávání nejlepší cesty ve složeném grafu v operaci zvané dekódování. Využití koherentního teoretického rámce garantuje, že výsledná struktura bude optimální podle definovaného kritéria. WFST mohou být v rámci daného polookruhu (semi-ring) optimalizovány pomocí determinizace a minimalizace. Aplikací těchto algoritmů získáme optimální strukturu pro prohledávání, optimální distribuce vah je pak získána aplikací "weight pushing" algoritmu. Cílem této práce je zdokonalit postupy a algoritmy pro konstrukci optimálních rozpoznávacích sítí. Zavádíme alternativní weight pushing algoritmus, který je vhodný pro důležitou třídu modelů -- převodníky jazykového modelu (language model transducers) a obecně pro všechny cyklické WFST a WFST se záložními (back-off) přechody. Představujeme také způsob konstrukce rozpoznávací sítě vhodné pro dekódování zpětně v čase, které prokazatelně produkuje ty samé pravděpodobnosti jako dopředná síť. K tomuto účelu jsme vyvinuli algoritmus pro exaktní reverzi back-off jazykových modelů a převodníků, které je reprezentují. Pomocí zpětných rozpoznávacích sítí optimalizujeme dekódování: ve statickém dekodéru je využíváme pro dvoustupňové dekódování (dopředné a zpětné vyhledávání). Tento přístup --- "sledovací" dekódování (tracked decoding) --- umožnuje zahrnout výsledky vyhledávání z prvního stupně do druhého stupně tak, že se sledují hypotézy obsažené v rozpoznávacím grafu (lattice) prvního stupně. Výsledkem je podstatné zrychlení dekódování, protože tato technika umožnuje prohledávat s  variabilním prohledávacím paprskem (search beam) -- ten je povětšinou mnohem užší než u základního přístupu. Ukazujeme rovněž, že uvedenou techniku je možné využít v dynamickém dekodéru tím, že postupně zjemňujeme rozpoznávání. To navíc vede i k částečné paralelizaci dekódování.

Tato práce se zabývá vyhledáváním výrazů v řeči pomocí mluvených příkladů (QbE STD). Výrazy jsou zadávány v mluvené podobě a jsou vyhledány v množině řečových nahrávek, výstupem vyhledávání je seznam detekcí s jejich skóre a časováním. V práci popisujeme, analyzujeme a srovnáváme tři různé přístupy ke QbE STD v jazykově závislých a jazykově nezávislých podmínkách, s jedním a pěti příklady na dotaz. Pro naše experimenty jsme použili česká, maďarská, anglická a arabská (levantská) data, a pro každý z těchto jazyků jsme natrénovali 3-stavový fonémový rozpoznávač. To nám dalo 16 možných kombinací jazyka pro vyhodnocení a jazyka na kterém byl natrénovaný rozpoznávač. Čtyři kombinace byly tedy závislé na jazyce (language-dependent) a 12 bylo jazykově nezávislých (language-independent). Všechny QbE systémy byly vyhodnoceny na stejných datech a stejných fonémových posteriorních příznacích, pomocí metrik: nesdružené Figure-of-Merit (non pooled FOM) a námi navrhnuté nesdružené Figure-of-Merit se simulací normalizace přes promluvy (utterrance-normalized non-pooled Figure-of-Merit). Ty nám poskytly relevantní údaje pro porovnání těchto QbE přístupů a pro získání lepšího vhledu do jejich chování. QbE přístupy použité v této práci jsou: sekvenční statistické modelování (GMM/HMM), srovnávání vzorů v příznacích (DTW) a srovnávání grafů hypotéz (WFST). Abychom porovnali výsledky QbE přístupů s běžnými STD systémy vyhledávajícími textové výrazy, vyhodnotili jsme jazykově závislé konfigurace také s akustickým detektorem klíčových slov (AKWS) a systémem pro vyhledávání fonémových řetězců v grafech hypotéz (WFSTlat). Jádrem této práce je vývoj, analýza a zlepšení systému WFST QbE STD, který po zlepšení dosahuje podobných výsledků jako DTW systém v jazykově závislých podmínkách.

No description available.

Indiana University-Purdue University Indianapolis (IUPUI) / A wave rotor constant volume combustor (WRCVC) was designed and built as a collaborative work of Rolls Royce LibertyWorks, Indiana University-Purdue University at Indianapolis (IUPUI), and Purdue University, and ran experimental tests at Purdue's Zucrow Laboratories in 2009. Instrumentation of the WRCVC rig inlet flow included temperature and pressure transducers upstream of the venturi and at the fuel delivery plane. Other instrumentation included exhaust pressures and temperatures. In addition, ion sensors, dynamic pressure sensors, and accelerometers were used to instrument the rotating hardware. The rig hardware included inlet guide vanes directly in front of the rotating hardware, which together with concern for damage potential, prevented use of any pressure transducers at the entrance to the rotor. For this reason, a complete understanding of the conditions at the WRCVC inlet is unavailable, requiring simulations of the WRCVC to estimate the inlet pressure at a specific operating condition based on airflow. The operation of a WRCVC rig test is a sequence of events over a short time span. These events include introduction of the main air flow followed by time-sequenced delivery of fuel, lighting of the ignition source, and the combustion sequence. The fast changing conditions in the rig inlet hardware make necessary a time-dependent computation of the rig inlet section in order to simulate the overall rig operation. The chosen method for computing inlet section temperature and pressure was a time-dependent lumped volume model of the inlet section hardware, using a finite difference modified Euler predictor-corrector method for computing the continuity and energy equations. This is coupled with perfect gas prediction of venturi air and fuel flow rates, pressure drag losses at the fuel nozzles, pressure losses by mass addition of the fuel or nitrogen purge, friction losses at the inlet guide vanes, and a correlation of the non-dimensional flow characteristics of the WRCVC. The flow characteristics of the WRCVC are computed by varying the non-dimensional inlet stagnation pressure and the WRCVC's operational conditions, assuming constant rotational speed and inlet stagnation temperature. This thesis documents the creation of a computer simulation of the entire WRCVC rig, to understand the pressure losses in the inlet system and the dynamic coupling of the inlet section and the WRCVC, so that an accurate prediction of the WRCVC rotor inlet conditions can be computed. This includes the computational development of the WRCVC upstream rig dynamic model, the background behind supporting computations, and results for one test sequence. The computations provide a clear explanation of why the pressures at the rotor inlet differ so much from the upstream measured values. The pressure losses correlate very well with the computer predictions and the dynamic response tracks well with the estimation of measured airflow. A simple Fortran language computer program listing is included, which students can use to simulate charging or discharging of a container.

With the development of intelligent era, speech recognition has been a hottopic. Although many automatic speech recognition(ASR) tools have beenput into the market, a considerable number of them do not support Swedishbecause of its small number. In this project, a Swedish ASR model basedon Hidden Markov Model and Gaussian Mixture Models is established usingKaldi which aims to help ICA Banken complete the classification of aftersalesvoice calls. A variety of model patterns have been explored, whichhave different phoneme combination methods and eigenvalue extraction andprocessing methods. Word Error Rate and Real Time Factor are selectedas evaluation criteria to compare the recognition accuracy and speed ofthe models. As far as large vocabulary continuous speech recognition isconcerned, triphone is much better than monophone. Adding feature transformationwill further improve the speed of accuracy. The combination oflinear discriminant analysis, maximum likelihood linear transformand speakeradaptive training obtains the best performance in this implementation. Fordifferent feature extraction methods, mel-frequency cepstral coefficient ismore conducive to obtain higher accuracy, while perceptual linear predictivetends to improve the overall speed. / Det existerar flera lösningar för automatisk transkribering på marknaden, menen stor del av dem stödjer inte svenska på grund utav det relativt få antalettalare. I det här projektet så skapades automatisk transkribering för svenskamed Hidden Markov models och Gaussian mixture models genom att användaKaldi. Detta för att kunna möjliggöra för ICABanken att klassificera samtal tillsin kundtjänst. En mängd av modellvariationer med olika fonemkombinationsmetoder,egenvärdesberäkning och databearbetningsmetoder har utforskats.Word error rate och real time factor är valda som utvärderingskriterier föratt jämföra precisionen och hastigheten mellan modellerna. När det kommertill kontinuerlig transkribering för ett stort ordförråd så resulterar triphonei mycket bättre prestanda än monophone. Med hjälp utav transformationerså förbättras både precisionen och hastigheten. Kombinationen av lineardiscriminatn analysis, maximum likelihood linear transformering och speakeradaptive träning resulterar i den bästa prestandan i denna implementation.För olika egenskapsextraktioner så bidrar mel-frequency cepstral koefficiententill en bättre precision medan perceptual linear predictive tenderar att ökahastigheten.

[pt] Ondas ultrassônicas do tipo SH (do inglês, shear horizontal), ou de cisalhamento horizontal são úteis no campo de ensaios não-destrutivos. Essas podem ser geradas, em meios metálicos, por transdutores acústicos eletromagnéticos (EMAT) que utilizam um arranjo de ímãs permanentes periódicos (PPM), chamados de EMATs PPM. Tais dispositivos em sua forma convencional, entretanto, geram ondas ultrassônicas bidirecionalmente, ou seja, para frente e para trás. Essa característica é geralmente indesejada. Recentemente, foi proposto um EMAT tipo PPM duplo com separação lateral que gera ondas nominalmente em uma única direção. Entretanto, seu diagrama de radiação revela lóbulos traseiros laterais. Esta dissertação tem como objetivo aprimorar o supracitado projeto através de seus parâmetros de construção, mais especificamente, o número de fileiras de ímãs permanentes do arranjo PPM e a separação lateral entre as mesmas. Foi desenvolvido um modelo analítico, baseado em linhas de fontes lineares, que considera a projeção de cada espira como uma fonte harmônica independente, para cálculo do campo ultrassônico gerado por EMATs PPM, tanto os do tipo PPM duplo, quanto o convencional. Com isso, foi possível obter o digrama de radiação teórico dos dispositivos, de forma a analisar como os seus parâmetros de construção interferem em sua unidirecionalidade. Utilizou-se três tecnologias distintas de bobinas, a saber, bobinas fabricadas manualmente, bobinas fabricadas a partir de placas de circuito impresso (PCB) flexível de poliamida e PCB flexível de poliéster. Os arranjos PPM duplos foram colocados sobre moldes impressos em impressora 3D. Realizou-se testes experimentais em uma placa de alumínio 1.5 mm de espessura, com EMAT PPM comercial utilizado como receptor, em diversos ângulos, com o objetivo de calcular o diagrama de radiação experimental do EMAT PPM duplo. Ao todo foram montadas 44 unidades quando combinadas todas as tecnologias de fabricação de bobina, com separação lateral de 1 mm a 4 mm e 1 a 4 de fileiras de ímãs. Resultados experimentais e teóricos mostraram boa concordância. Foi observado que tanto aumentar o número de fileiras de ímãs, quanto diminuir a sua separação lateral reduzem a intensidade dos lóbulos traseiros laterais. Resultados experimentais mostraram que, com a configuração de 4 fileiras de ímãs, separados lateralmente por 1 mm e bobinas feitas a partir de PCB de poliamida, a intensidade dos lóbulos traseiros laterais foi reduzida em 8.6 dB, em relação à configuração original do PPM EMAT duplo . Sobre as tecnologias de bobina, pode-se concluir que o uso de PCBs flexíveis tornou possível obter menores separações laterais entre as fileiras de ímãs e simplificar o processo de construção do EMAT PPM duplo, com destaque para a bobina feita com PCB flexível de poliamida, que apresentou baixo erro médio entre medidas experimentais e simuladas, apresentando também o mais fácil processo de fabricação entre os três tipos de bobinas avaliados. / [en] Shear horizontal (SH) ultrasonic waves are useful in the non-destructive evaluation field. They can be generated, in metallic media, by electromagnetic acoustic transducers (EMAT) that use an array of periodic permanent magnets (PPM). PPM EMATs, in their conventional form, generate forward and backward traveling SH waves. This feature is generally undesired. Recently, a side-shifted dual-PPM EMAT has been proposed. This device generates waves predominantly in a single direction. However, its radiation pattern reveals backward side lobes. This thesis aims to improve the aforementioned design through its construction parameters, more specifically, the number of rows of magnets in the PPM array and the lateral separation between them. An analytical model, based on the line source method, which considers the wires projections as independent wave sources, was developed to calculate the ultrasonic field generated by PPM EMATs, either for the conventional one or the dual-PPM EMAT. Exploiting the developed model, it was possible to compute the theoretical radiation pattern of the device in order to better analyse how its construction parameters affect its unidirectionality. Three different coil technologies were used, namely hand-wound coils and coils made with flexible printed circuit boards (PCBs), either of polyamide or polyester material. The magnet arrays were mounted on 3D-printed molds to ensure their positioning. Experimental tests were carried out where the fabricated device was placed onto a 1.5 mm-thick aluminum plate, and a commercial PPM EMAT was used as a receiver, at different reception angles, in order to calculate the radiation pattern of the dual-PPM EMAT. Overall, 44 units, combining all coil technologies, were manufactured with lateral separation from 1 mm to 4 mm and 1 to 4 rows of magnets per PPM. Experiments and theoretical results presented good agreement. It was observed that either increasing the number of rows of magnets or decreasing their lateral separation reduced the intensity of the backward side lobes. When using the configuration of 4 rows of magnets and 1 mm side shift, fabricated with polyamide PCBs, results revealed that the intensity of the backward side lobes was reduced by 8.6 dB, when compared with the original configuration. Regarding the coil technologies, it can be concluded that the use of flexible PCBs allowed one to obtain shorter lateral separations between rows of magnets and simplified the fabrication process of the dual-EMAT. Mainly, the poliamide-based flexible PCB coil presented low average error between experimental and simulated measurements and the easiest fabrication process among all three coil technologies tested.

<p dir="ltr">Rapid evolution in wireless technology and the increasing demand for high bandwidth communication for 5G/6G and the Internet of Things (IoT) have necessitated a growing number of components in radio front-end modules in an increasingly overcrowded radio frequency (RF) spectrum. Low-cost ad-hoc radios have drawn consumer interest, enabling new devices like microelectromechanical (MEMS) resonators for on-chip clocking, frequency-selective filters, RF signal processing, and spectral sensing for their small footprint and low power consumption. Gallium nitride (GaN) is an attractive electromechanical material due to its high coupling coefficient, acoustic velocity, and low viscoelastic losses. These benefits enable high-Q MEMS resonators in GaN monolithic microwave integrated circuits (MMICs) with scaling capability up to mm-wave frequencies, making this technology platform a contender for high-performance programmable radios in RF/mm-wave, sensors for harsh environments, and information processing in quantum systems.</p><p dir="ltr">The bias-dependent control mechanism of the 2D electron gas (2DEG) in GaN heterostructures can be exploited to design different switchable and tunable devices for reconfigurable MEMS components. This work presents, for the first time, a comprehensive study of the electromechanical performances of different transduction mechanisms in switchable GaN MEMS resonators. A unique OFF-state shunt design, where the 2DEG in an AlN/GaN heterostructure is utilized to control electromechanical transduction in Lamb mode resonators, is also experimentally demonstrated in this work. To make a valid comparison among switchable transducers, equivalent circuit models are developed to extract key parameters from the measurements by fitting them in both ON and OFF states. The switchable transducer with Ohmic interdigitated transducers (IDTs) and Schottky control gate shows superior performance among the designs under consideration with complete suppression of the mechanical mode in the OFF state and a maximum frequency-quality factor product of 5x10¹²s^-1 and a figure-of-merit of 5.18 at 1GHz in the ON state.</p><p dir="ltr">Over the past few years, there have been numerous efforts to scale the frequencies of MEMS devices in the GaN platform towards mm-wave frequencies. However, challenges remain due to the multi-layer thick buffer, typical in the growth of GaN epilayer on a substrate. This work presents the investigation of SweGaN QuanFINEbuffer-free and ultrathin GaN-on-SiC for the performance of surface acoustic wave (SAW) devices beyond 10GHz. Finite element analysis (FEA) is performed to find the range of frequencies for the Sezawa mode in the structure. Transmission lines and resonators are designed, fabricated, and characterized. Modified Mason circuit models are developed for each class of devices to extract critical performance metrics and benchmark with the state-of-the-art and theoretical limits for GaN. Sezawa modes are observed at frequencies up to 14.3GHz, achieving a record high in GaN MEMS to the best of our knowledge. A maximum piezoelectric coupling of 0.61% and frequency-quality factor product of 6x10¹²s^-1 are achieved for Sezawa resonators at 11GHz, with a minimum propagation loss of 0.26dB/λ for the two-port devices. The devices also exhibit high linearity with input third-order intercept points (IIP3) of 65dBm at 9GHz.</p><p dir="ltr">This work also investigates tunable acoustoelectric (AE) devices in the QuanFINE platform, leveraging its inherent 2DEG in the AlGaN/GaN heterostructure. Using 9.7GHz Sezawa mode acoustic delay lines, we report the highest frequency of AE in GaN to date. Active and passive AE devices are designed for voltage-dependent non-reciprocity and propagation loss without modification to the standard process for the High Electron Mobility Transistors (HEMTs) in MMICs. Drain/source Ohmic contacts control the drift velocity of the 2DEG, and the Schottky gate modulates 2DEG carrier concentration, resulting in a 30dB/cm separation between forward and reverse acoustic waves for a 2.56kV/cm lateral DC electric field and a maximum change in propagation loss of 50dB/cm for -5V DC at the control gate, respectively. The QuanFINEtechnology with AlGaN/GaN heterostructure enables a platform for switchable MEMS resonators and tunable acoustoelectric devices in MMICs for reconfigurable front end approaching mm-wave frequencies.</p>

Future robotic systems will be pervasive technologies operating autonomously in unknown spaces that are shared with humans. Such complex interactions make it compulsory for them to be lightweight, soft, and efficient in a way to guarantee safety, robustness and long-term operation. This set of qualities can be achieved using soft multipurpose systems that combine, integrate and commute between conventional electromechanical and fluidic drives, as well as harvest energy during inactive actuation phases for increased energy efficiency. Recent research work has shown that dielectric fluids with specific properties, can be combined with stretchable or flexible shell structures made of polymeric dielectric/electrode composite films, to implement a novel type of soft electrically-driven fluidic transducers with self-healing and self-sensing capabilities that take the name of Liquid-based Electro-Active Polymer transducers (LEAPs). These devices are similar to dielectric elastomer transducers in regards to their electrostatic working principle, but they can potentially produce larger displacements due to their lower mechanical stiffness. In this thesis a novel electrostatic transducer is presented; the transducer is made of thin polymer films and liquid dielectrics, combined with rigid stiffening elements to form a circular electrostatic bellow muscle (EBM) unit capable of out-of-plane contraction. These units are easy to manufacture and can be arranged in arrays and stacks that can be employed as contractile artificial muscles, pumps for fluid-driven soft robots, or as energy harvesters. As artificial muscles, EBMs of 20 - 40 millimeters in diameter can exert forces of up to 6 newtons, lift loads over a hundred times their own weight, and reach contractions of over 40 per cent with strain rates over 1200 per cents per second, with a bandwidth over 10 Hz. As pump drivers, EBMs produce flow rates of up 0.63 liters per minute and maximum pressure head of 6 kilopascals, whereas as generators, they reach a conversion efficiency close to 20 per cent. The compact shape, low cost, simple assembling procedure, high reliability and large contractions make the EBM a promising technology for high-performance robotic systems.

Esta tesis reune algunas contribuciones al reconocimiento de formas estadístico y, más especícamente, a varias tareas del procesamiento del lenguaje natural. Varias técnicas estadísticas bien conocidas se revisan en esta tesis, a saber: estimación paramétrica, diseño de la función de pérdida y modelado estadístico. Estas técnicas se aplican a varias tareas del procesamiento del lenguajes natural tales como clasicación de documentos, modelado del lenguaje natural y traducción automática estadística. En relación con la estimación paramétrica, abordamos el problema del suavizado proponiendo una nueva técnica de estimación por máxima verosimilitud con dominio restringido (CDMLEa ). La técnica CDMLE evita la necesidad de la etapa de suavizado que propicia la pérdida de las propiedades del estimador máximo verosímil. Esta técnica se aplica a clasicación de documentos mediante el clasificador Naive Bayes. Más tarde, la técnica CDMLE se extiende a la estimación por máxima verosimilitud por leaving-one-out aplicandola al suavizado de modelos de lenguaje. Los resultados obtenidos en varias tareas de modelado del lenguaje natural, muestran una mejora en términos de perplejidad. En a la función de pérdida, se estudia cuidadosamente el diseño de funciones de pérdida diferentes a la 0-1. El estudio se centra en aquellas funciones de pérdida que reteniendo una complejidad de decodificación similar a la función 0-1, proporcionan una mayor flexibilidad. Analizamos y presentamos varias funciones de pérdida en varias tareas de traducción automática y con varios modelos de traducción. También, analizamos algunas reglas de traducción que destacan por causas prácticas tales como la regla de traducción directa; y, así mismo, profundizamos en la comprensión de los modelos log-lineares, que son de hecho, casos particulares de funciones de pérdida. Finalmente, se proponen varios modelos de traducción monótonos basados en técnicas de modelado estadístico . / Andrés Ferrer, J. (2010). Statistical approaches for natural language modelling and monotone statistical machine translation [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/7109

Optical current sensors (OCSs) are initially developed to measure relatively large current over a wide range of frequency band. They are also used as protective devices in the event a fault occurs due to a short circuit, in the power generation and distribution industries. The basic principal used in OCS is the Faraday effect. When a light guiding faraday medium is placed in a magnetic field which is produced by the current flowing in the conductor around the magnetic core, the plane of polarization of the linearly polarized light is rotated. The angle of rotation is proportional to the magnetic field strength, proportionality constant and the interaction length. The proportionality constant is the Verdet constant V (λ, T), which is dependent on both temperature and wavelength of the light. Opto electrical methods are used to measure the angle of rotation of the polarization plane. By measuring the angle the current flowing in the current carrying conductor can be calculated. But the accuracy of the OCS is lost of the angle of rotation of the polarization plane is dependent on the Verdet constant, apart from the magnetic field strength. As temperature increases the Verdet constant decreases, so the angle of rotation decreases. To compensate the effect of temperature on the OCS, a new method has been proposed. The current and temperature are measured with the help of a duel frequency method. To detect the line current in the conductor or coil, a small signal from the line current is fed to the reference of the lock in. To detect the temperature, the coil is excited with an electrical signal of a frequency different from the line frequency, and a small sample of this frequency signal is applied to the reference of the lock in. The temperature and current readings obtained are look up at the database value to give the actual output. Controlled environment is maintained to record the values in the database that maps the current and temperature magnitude values at the DSP lock in amplifier, to the actual temperature and current. By this method we can achieve better compensation to the temperature changes, with a large dynamic range and better sensitivity and accuracy.