The topic of this thesis is to implement efficient decoder for speech recognition training system ASR Kaldi (http://kaldi.sourceforge.net/). Kaldi is already deployed with decoders, but they are not convenient for dialogue systems. The main goal of this thesis to develop a real time decoder for a dialogue system, which minimize latency and optimize speed. Methods used for speeding up the decoder are not limited to multi-threading decoding or usage of GPU cards for general computations. Part of this work is devoted to training an acoustic model and also testing it in the "Vystadial" dialogue system. Powered by TCPDF (www.tcpdf.org)

One of the greatest challenges of human today is to provide the continuous and sustainable energy supply to the worldwide society. This shall be done while minimizing all the negative consequences of the operation(s) to the environment and its living habitants including human beings, taking from the whole life cycle perspective. In this thesis work new solutions for treatment biomass and waste are analyzed.   Based on the fundamental research on the conversion of various materials (biomass: straw pellets, wood pellets; and waste: plastic waste, ASR residues after pyrolysis), converted by means of different systems (pyrolysis in a fluidized bed reactor, gasification in a fixed-bed reactor using highly preheated agents) it is recommended to classify materials against their charring properties under pyrolysis, in order to find the best destination for a given type of fuel.    Based on phenomenological research it was found that one of the important effects, affecting performance of downdraft gasifiers, is the pressure drop through the bed and grate. It affects, directly, the velocity profile, temperature distribution and of the height of the bed, especially for the grate with restricted passage surface, although it was not investigated in literature. The lower grate porosity, the higher conversion of fuel and heating value of gas is produced. However, the stability of the process is disturbed; therefore reducing the grate porosity below 20% is not recommended, unless the system is designed to overtake the consequences of the rising pressure inside the reactor. This work proposed the method for prediction of a total pressure drop through the fixed-bed downdraft gasifier equipped with a grate of certain porosity with an uncertainty of prediction ±7.10.     Three systems have been proposed; one for the treatment of automotive shredder residue (ASR), one for the treatment of plastic waste (polyolefins) and one for biomass (wood/straw pellets). Pyrolysis is an attractive mean of conversion of non-charring materials (like plastic waste) into valuable hydrocarbons feedstock. It gives directly 15-30% gaseous olefins while the residue consisting of naphtha-like feedstock has to be reformed/upgraded to olefins or other chemicals (e.g. gasoline generation) using available petrochemical technologies. Pyrolysis of complex waste mixture such as ASR is an attractive waste pretreatment method before applying any further treatments, whereby useful products are generated (gaseous and liquid fuel) and char, rich in precious metals. The solid residues are meant for further treatment for energy and metals recovery. Gasification is a complementary method for handling pyrolysis residues. However, metals can be removed before gasification. Pyrolysis of charring materials, like biomass, is a very important step in thermo-chemical conversion. However, the char being approximately 25%wt. contains still very high caloric value of about 30MJ/kg. This in connection with the High Temperature Steam Gasification process is a very promising technology for biomass treatment, especially, above 900oC. This enhances the heat transfer towards the sample and accelerates kinetics of the gasification. This, in turn, improves the conversion of carbon to gas, increases the yield of the producer gas and reduces tar content. At higher steam to fuel ratio the process increases the yield of hydrogen, making it suitable for second-generation biofuels synthesis, whereas at lower steam to fuel ratio (S/F&lt;2) the generated gas is of high calorific value making it suitable for power generation in a combined cycle. / <p>QC 20110607</p>

Um grande empecilho para a manutenção e estabilidade da produção nacional de soja reside na suscetibilidade dos diferentes genótipos aos estresses ambientais. Tendo em vista a importância social e econômica da leguminosa e os efeitos extremamente danosos dos estresses abióticos sobre a agricultura, faz-se necessário maior conhecimento acerca das interações entre os estímulos estressores e as respostas da planta. A seca é considerada o principal fator limitante na produtividade agrícola. Sendo assim, a identificação e caracterização de genes responsivos a essa condição é um passo inicial na compreensão das respostas adaptativas ao déficit hídrico. Os genes Asr (ABA, Stress and Ripening) são induzidos por estresse e ácido abscísico (ABA) e seus níveis de expressão são rapidamente aumentados em resposta à salinidade e seca. Nesse estudo os genes da família Asr de soja foram clonados. Estas proteínas são hidrofílicas e ricas nos aminoácidos Ala, His, Glu e Lis, apresentando homologia com ASRs de outras plantas, como atestado nas análises de múltiplos alinhamentos. O perfil de expressão foi avaliado através de RT-qPCR em tempo real e revelou que Asr1 tem um distinto padrão de indução no nível de transcritos em folha sob tratamento com ABA, sal e seca, enquanto Asr3 apresenta padrão distinto de indução na expressão em raiz, sob tratamento com ABA e seca. Além disto, foram construídos vetores para a superexpressão e localização subcelular das proteínas ASR1, ASR2 e ASR3 em plantas. Plantas de Arabidopsis thaliana foram submetidas a um protocolo de transformação genética mediada por Agrobacterium. / One of the major obstacles to maintain the stability of the national production of soybean (Glycine max) lies on the susceptibility of different genotypes to abiotic stress. In view of the social and economic importance of soybean and due to the extremely harmful effects of stress in agriculture, detailed knowledge of the interaction between these stresses and plant response to environmental stimuli is necessary. Drought is considered the main abiotic limitation factor for agricultural productivity. Identification and characterization of responsive genes to this condition is an initial step in understanding the adaptive responses to drought. The Asr (ABA, Stress and Ripening) genes are induced by stress and abscisic acid (ABA) in plants, and their expression levels are quickly increased in response to salinity and drought. In this study Asr genes from Glycine max were cloned. These proteins were found to be hydrophilic and rich in amino acids Ala, His, Glu and Lys, showing homology with those of other plant Asr genes via multiple alignment analysis. RT-qPCR analyses revealed that Asr1 had a distinct up-regulated transcript pattern in leaf under ABA, NaCl and drought treatments, while Asr3 had a distinct up-regulated transcript pattern in root under ABA and drought treatments. Besides, vectors for ASR1, ASR2 and ASR3 proteins overexpression and subcellular localization in plants were constructed. Arabidopsis thaliana plants were submitted to an Agrobacterium-mediated transformation procedure.

Um dos graves obstáculos para a manutenção e estabilidade da produção nacional de arroz (Oryza sativa) reside na susceptibilidade dos genótipos existentes a estresses abióticos. Tendo em vista a importância social e econômica do arroz e os efeitos extremamente danosos desses estresses sobre a agricultura, o conhecimento detalhado das interações entre os estresses abióticos e as respostas dos vegetais frente a esses estímulos ambientais faz-se necessário. O alumínio (Al) é considerado um dos principais fatores limitantes na produção agrícola, inibindo o crescimento das raízes e a absorção de minerais. A toxicidade do Al em plantas ocorre pela sua solubilização em solos com pH baixo ou solos ácidos. Os genes ASR (ABA, Stress and Ripening) são induzidos por estresse e ácido abscísico (ABA) em plantas, e seus níveis de expressão são rapidamente aumentados em resposta à salinidade e seca. Recentemente, foi demonstrado que o gene que codifica a proteína ASR5 é responsivo ao Al em raízes de arroz. Apesar do arroz ser considerado um dos cereais mais resistentes a Al, os mecanismos básicos de tolerância a este metal são pouco conhecidos no arroz em comparação a outros cereais. Por meio do presente trabalho objetivamos: i) a caracterização funcional dos membros da família gênica ASR de arroz em reposta ao Al; e ii) a construção de vetores binários de transformação de plantas visando o estudo da localização subcelular da proteína codificada pelo gene OsASR5, e o silenciamento gênico da família ASR de arroz. As análises dos transcritos por qRT-PCR mostraram que todos os genes da família ASR de arroz ssp Japonica respondem ao Al. Por outro lado, OsASR5 não sofre modulação de sua expressão em resposta ao Al em raízes de arroz ssp Indica. Essas diferenças de respostas dos genes OsASR5 em distintas variedades pode refletir diferenças no grau de tolerância ao Al de cada um desses genótipos. / One of the major obstacles to maintain the stability of the national production of rice (Oryza sativa) lies on the susceptibility of the different genotypes to abiotic stress. In view of the social and economic importance of rice and due to the extremely harmful effects of stress in agriculture, detailed knowledge of the interactions between these stresses and plant responses to environmental stimuli is necessary. Aluminum (Al) is considered one of the main limitation factors for agricultural productivity, inhibiting root growth and mineral absorption. Al toxicity occurs by its solubilization in soils with low pH or acid soils. The ASR (ABA, Stress and Ripening) genes are induced by stress and abscisic acid (ABA) in plants, and their expression levels are quickly increased in response to salinity and drought. Recently, it was demonstrated that the ASR5 gene is responsive to Al in rice roots. Despite the fact that rice is considered one of the most resistant crops to Al, the basic mechanisms of tolerance to this metal are poorly known when compared to other crops. This study aimed the functional characterization of the gene expression of rice ASR family members in response to Al, and the construction of binary vectors for the subcellular localization of the protein codified by the OsASR5 gene, and the construction of a gene silencing binary vector for the ASR family. Analyses of transcripts by qRT-PCR showed that in the ssp Japonica, all ASR genes responded to Al. In contrast, OsASR5 do not suffer expression modulation in response to Al in rice roots of ssp Indica. These differences in response of the OsASR5 gene in distinct varieties may reflect differences in the degree of Al tolerance in each genotype.

För att bibehålla elnätet stabilt är det viktigt för elproducenterna attkunna möta samhällets behov av elkraft. Detta behov varierar beroendepå tid på dygnet och även av årstid. Att kunna samla energi då behovetär lågt, för att sedan utvinna och distribuera energi då behovet ökar ärdärför viktigt. Vattenkraft är en av de energikällor som är enklast attreglera. Denna energi är dessutom relativt miljövänlig. Att ha en stabiloch säker vattenkraft är därför viktigt för samhället.I detta examensarbete har vissa problem som kan uppstå i ettvattenkraftverkets studerats, närmare bestämt alkali-silikatreaktion ibetong. Denna reaktion framträder genom att betongen sväller. Tillföljd av detta kan konstruktionen spricka. Detta beror på att en gelbildas när alkalier och kisel reagerar med varandra. Denna gel kan taupp vatten och då svälla.En specifik vattenkraftstation har i detta examensarbete studeratsnärmare, nämligen Malgomaj kraftverk. Detta är en anläggning somligger i ett område där, till skillnad från övriga Sverige, det finnsbergarter som har en medelsnabb reaktion med avseende på alkalikiselreaktion.Att denna geografiska placering blir ett problem beror påatt det stenmaterial som finns att tillgå i vattenkraftstationens närhethar använts som ballast i anläggningens betongkonstruktion.I den vattenkraftstation som studeras har problem iakttagits på grundav svällningar av betongkonstruktionen kring turbinen. För att få enuppskattning om hur vattenkraftstationens deformationer i framtidenkommer att utbildas har en modell av problemområdet byggts uppmed hjälp av finita elementmetoden, en så kallad FEM-modell. Dennamodell kalibreras mot mätdata och ska sedan ligga till grund för enuppskattning av vattenkraftstationens livslängd.Resultatet från undersökningen i detta examensarbete visar attdeformationerna är små men betydande för vattenkraftstationensmöjlighet till att fortsätta sin energiproduktion. / To maintain a stable power grid, it is important for electricity producers to meetsociety's need for electricity. This need will vary depending on time of day and eventhe season. Being able to accumulate energy when demand is low, and regain energywhen demand increases, is therefore important. Hydropower is one of the energysources that are easiest to regulate. Having a stable and secure hydropower istherefore important for society.In this thesis one problem that can occur in a hydroelectric plant has been studied,namely alkali-silica reaction (ASR) in concrete. This reaction causes the concrete toswell, due to a formation of gel when alkali and silicon react together.A specific hydropower station has been studied in detail, namely Malgomajhydropower plant. This is a facility that is located in an area where, unlike the rest ofSweden, there are stone materials that have a moderately rapid reaction with respectto the ASR.Problems for this hydroelectric power station have been observed because of swellingof the concrete structure surrounding the turbine. To get an estimate prognosis ofhow the hydropower plant will deform in the future, a finite element method-model(FEM-model) has be created of the problem area. This model is calibrated againstmeasured data and will then form the basis for an appreciation of the hydropowerstation's remaining lifetime.The results in this thesis show that the deformations are small but significant for thehydropower station's opportunity to continue its energy production.

Conventional speech recognizers employ a training phase during which many of their parameters are configured - including vocabulary selection, feature selection, and decision mechanism tailoring to these selections. After this stage during normal operation, these traditional recognizers do not significantly alter any of these parameters. Conversely this work draws heavily on high level human thought patterns and speech perception to outline a set of precepts to eliminate this training phase and instead opt to perform all its tasks during the normal operation. A feature space model is discussed to establish a set of necessary and sufficient conditions to guide real-time feature selection. Detailed implementation and preliminary results are also discussed. These results indicate that benefits of this approach can be seen in increased speech recognizer adaptability while still retaining competitive recognition rates in controlled environments. Thus this can accommodate such changes as varying vocabularies, class migration, and new speakers.

Conventional speech recognizers employ a training phase during which many of their parameters are configured - including vocabulary selection, feature selection, and decision mechanism tailoring to these selections. After this stage during normal operation, these traditional recognizers do not significantly alter any of these parameters. Conversely this work draws heavily on high level human thought patterns and speech perception to outline a set of precepts to eliminate this training phase and instead opt to perform all its tasks during the normal operation. A feature space model is discussed to establish a set of necessary and sufficient conditions to guide real-time feature selection. Detailed implementation and preliminary results are also discussed. These results indicate that benefits of this approach can be seen in increased speech recognizer adaptability while still retaining competitive recognition rates in controlled environments. Thus this can accommodate such changes as varying vocabularies, class migration, and new speakers.

The human auditory perception system is much more noise-robust than any state-of theart automatic speech recognition (ASR) system. It is expected that the noise-robustness of speech feature vectors may be improved by employing more human auditory functions in the feature extraction procedure. Forward masking is a phenomenon of human auditory perception, that a weaker sound is masked by the preceding stronger masker. In this work, two human auditory mechanisms, synaptic adaptation and temporal integration are implemented by filter functions and incorporated to model forward masking into MFCC feature extraction. A filter optimization algorithm is proposed to optimize the filter parameters. The performance of the proposed method is evaluated on Aurora 3 corpus, and the procedure of training/testing follows the standard setting provided by the Aurora 3 task. The synaptic adaptation filter achieves relative improvements of 16.6% over the baseline. The temporal integration and modified temporal integration filter achieve relative improvements of 21.6% and 22.5% respectively. The combination of synaptic adaptation with each of temporal integration filters results in further improvements of 26.3% and 25.5%. Applying the filter optimization improves the synaptic adaptation filter and two temporal integration filters, results in the 18.4%, 25.2%, 22.6% improvements respectively. The performance of the combined-filters models are also improved, the relative improvement are 26.9% and 26.3%.

Heterogeneities in concrete caused by the random distribution of aggregate in the cement-paste matrix lead to strong scattering of ultrasound waves at wavelengths on the order of the aggregate. Use of these high frequencies is necessary to detect damage at an early stage, something that is not possible with conventional ultrasonic methods. The ultrasound energy density in this regime can be described by the diffusion equation. The objective of this research is to develop a quantitative understanding of the effects of additional scattering sources, such as small cracks in the cement-paste matrix, on the parameters of the diffusion equation; these parameters are the diffusion and the dissipation coefficients. Experimentally measured ultrasonic waves are processed using the diffusion theory to determine the diffusivity and the dissipation coefficients as a function of frequency. The samples employed are made of a Portland cement-paste matrix and regular aggregate such as gravel and sand. The results provide a basic understanding of the repeatability and consistency of diffusion measurements, with an emphasis on the nondestructive evaluation of damage in concrete. In addition, a method to describe concrete in the coherent regime is examined. Existing wave propagation models for inhomogeneous materials deal with two-phase mixtures, mostly the matrix-inclusion system such as fiber-reinforced composites. There are, however, numerous examples of multi-phase materials in which more than one phase is suspended in a matrix-phase. This research considers concrete, in which cement paste and aggregates with different sizes and mechanical properties are mixed together. Most of the models for two-phase composites cannot be extended to a multi-phase composite. Among others, the effective medium theory is considered here for two reasons: first, the formalism in this theory can easily be extended to multi-phase cases; second, the theory does not strictly define a specific microstructure between phases, which allows for a simulation of the microstructure in which different inclusions are in contact. The mathematical formulation is presented that yields the formulae for the effective density and the effective bulk and shear moduli. Finally, the calculated wave speeds and attenuations for different materials are compared with experimental results.

Aquifer Storage Recovery (ASR) is a recently developed (circa 1970) method (in the U.S.A.) to reduce groundwater-pumping stresses by injecting treated wastewater or surface water during periods of low demand into an aquifer followed by its recovery during periods of high demand. This method has been successfully implemented in numerous locations across the U.S.A. and worldwide, mainly due to the cost savings provided by the use of an unlimited reservoir (aquifer) in which to store water compared to the costs to construct surface impoundments and the inherent problems with storing such water for extended periods of time under evaporative atmospheric conditions. "This study describes the use of a highly discretized, three-dimensional, variable-density, numerical model (SEAWAT 2000) that incorporates the vertical variation of hydraulic conductivities, measured foot by foot, from a continuous core collected from the upper Floridan aquifer in southwest Florida, to evaluate the effects of small-scale heterogeneities on a hypothetical ASR system well. In order to compare these effects to the more general case in which average hydraulic parameters are used to characterize flow zones, a model is constructed with average parameters taken from the heterogeneous case. This study attempts to determine whether aquifer heterogeneities influence the performance of ASR systems, compared to assumed homogeneous conditions, by quantifying differences in recovery efficiency, horizontal and vertical flow due to advection and dispersion, plume dimensions, and storage periods. The results of this study indicate that 1) the geometry of the injectate plume under homogeneous and heterogeneous conditions differ significantly; 2) background formation total dissolved solids (TDS) concentrations significantly control the quantity of potable water available for recovery; 3) dispersion exhibits a strong control on vertical mixing; 4) multiple injection cycles are required to generate a plume of potable water for long term storage; and 5) the percent recoveries under homogeneous and heterogeneous conditions are generally similar only in low-salinity background concentrations, due to the absence of the effects of buoyancy. Although the percent recoveries of the systems modeled are similar, the success of an ASR well is strongly controlled by the existence of heterogeneities, which essentially determine the degree of horizontal and vertical mixing of the injectate with formation waters. Heterogeneities result in varying groundwater and mass transport paths during injection and recovery periods. Presumably these variations would need to be considered when evaluating potential variations in groundwater quality due to mixing between formation and injected water. Understanding potential variations in groundwater quality and treatment alternatives due to the presence of ASR-associated geochemical conditions, e.g., elevated arsenic concentrations, may also be improved with a detailed heterogeneous numerical model.