• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 78
  • 35
  • 24
  • 13
  • 11
  • 8
  • 7
  • 6
  • 4
  • 3
  • 3
  • 2
  • 1
  • 1
  • Tagged with
  • 215
  • 38
  • 36
  • 32
  • 26
  • 25
  • 25
  • 23
  • 21
  • 21
  • 21
  • 20
  • 20
  • 19
  • 18
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
71

Speaker adaptation of deep neural network acoustic models using Gaussian mixture model framework in automatic speech recognition systems / Utilisation de modèles gaussiens pour l'adaptation au locuteur de réseaux de neurones profonds dans un contexte de modélisation acoustique pour la reconnaissance de la parole

Tomashenko, Natalia 01 December 2017 (has links)
Les différences entre conditions d'apprentissage et conditions de test peuvent considérablement dégrader la qualité des transcriptions produites par un système de reconnaissance automatique de la parole (RAP). L'adaptation est un moyen efficace pour réduire l'inadéquation entre les modèles du système et les données liées à un locuteur ou un canal acoustique particulier. Il existe deux types dominants de modèles acoustiques utilisés en RAP : les modèles de mélanges gaussiens (GMM) et les réseaux de neurones profonds (DNN). L'approche par modèles de Markov cachés (HMM) combinés à des GMM (GMM-HMM) a été l'une des techniques les plus utilisées dans les systèmes de RAP pendant de nombreuses décennies. Plusieurs techniques d'adaptation ont été développées pour ce type de modèles. Les modèles acoustiques combinant HMM et DNN (DNN-HMM) ont récemment permis de grandes avancées et surpassé les modèles GMM-HMM pour diverses tâches de RAP, mais l'adaptation au locuteur reste très difficile pour les modèles DNN-HMM. L'objectif principal de cette thèse est de développer une méthode de transfert efficace des algorithmes d'adaptation des modèles GMM aux modèles DNN. Une nouvelle approche pour l'adaptation au locuteur des modèles acoustiques de type DNN est proposée et étudiée : elle s'appuie sur l'utilisation de fonctions dérivées de GMM comme entrée d'un DNN. La technique proposée fournit un cadre général pour le transfert des algorithmes d'adaptation développés pour les GMM à l'adaptation des DNN. Elle est étudiée pour différents systèmes de RAP à l'état de l'art et s'avère efficace par rapport à d'autres techniques d'adaptation au locuteur, ainsi que complémentaire. / Differences between training and testing conditions may significantly degrade recognition accuracy in automatic speech recognition (ASR) systems. Adaptation is an efficient way to reduce the mismatch between models and data from a particular speaker or channel. There are two dominant types of acoustic models (AMs) used in ASR: Gaussian mixture models (GMMs) and deep neural networks (DNNs). The GMM hidden Markov model (GMM-HMM) approach has been one of the most common technique in ASR systems for many decades. Speaker adaptation is very effective for these AMs and various adaptation techniques have been developed for them. On the other hand, DNN-HMM AMs have recently achieved big advances and outperformed GMM-HMM models for various ASR tasks. However, speaker adaptation is still very challenging for these AMs. Many adaptation algorithms that work well for GMMs systems cannot be easily applied to DNNs because of the different nature of these models. The main purpose of this thesis is to develop a method for efficient transfer of adaptation algorithms from the GMM framework to DNN models. A novel approach for speaker adaptation of DNN AMs is proposed and investigated. The idea of this approach is based on using so-called GMM-derived features as input to a DNN. The proposed technique provides a general framework for transferring adaptation algorithms, developed for GMMs, to DNN adaptation. It is explored for various state-of-the-art ASR systems and is shown to be effective in comparison with other speaker adaptation techniques and complementary to them.
72

漲跌幅限制下股價行為與財務指標受扭曲程度之研究 / The Impacts of Stock Price Limits on Security Price Behavior and Financial Risk Indices Measures

黃健榮, Huang, Je Rome Unknown Date (has links)
我國股市的價格漲跌幅限制已逾三十年的歷史,主管機關維持此一機制的訴求是避免股價波動過於激烈、抑制投機行為。惟停板限制可能帶來的影響,除直覺上的其造成投資者持股風險指標扭曲等問題。經探究中亦歸結出(一)其被引為技術指標、(二)其引致財務風險指標扭曲等問題。   經探究GMM、Gibbs Sampler、與Two-Limit-Tobit Model模型的優劣。本研究發現一般使用的GMM估計量並非不偏,雖然可以藉修正增加其效率性,但仍無法藉以衡量各種的停板影響;Gibbs Sampler則過於依賴特定的先驗分佈,有可能因此而造成偏誤;而目前使用Tobit Model的文獻大都忽略停板限制對股價的影響力,據以產生的估計值亦附含偏誤。   本研究所採樣本期間為79年1月3日至84年10月9日,使用模型為Two-Limit-Tobit Model。為求嚴謹,在使用之前做資料的處理,並利用CAAR來驗證模型的正確性。實證顯示,漲跌停板的設立顯著改變投資人行為,在停板之前本研究發現存在技術指標與標準差統計量的向上偏誤,進而可能誤導實業界財務決策或學術研究結論。 / Thsi Study explores how price limits, which have remained in Taiwan Securities Exchange for over thirty years, affects both security price behavior and security risk indices. Its empirical results add to our understanding of the social costs and benefits of price limits. The SEC has been advocating the merits of price limits, emphasing that they help eliminating speculative trades and reducing security price volatility. In contrast, it remains a popular thought that price limits increase investors’holding costs and risks. To empirically examine the effects of price limits in Taiwan, this papers adopts Two-Limit-Tobit Model, together with CAAR as an indicator for specification validity. My test results lend support to the notion of (1).Technical Indicator Effect immediately before the price limits are hit; (2).Enhancement Effect the day after. Moreover, price limits contribute to bias in both systematic risk and total risk estimates (namely, β and σ) and thus distort investment decisions.   This Study also contributes to the contemporary literature by examining the merits and limitations of GMM, Gibbs Sampler, and Two-Limit-Tobit Model. GMM estimator is subject to statistical bias. One way may gain efficiency via adjustment. And yet GMM ahs pitfalls in directly measuring the price limit effects; The major limitation of the Gibbs Sampler is its reliance on specific prior information and it may lead to bias. And most of the papers adopting Tobit Model simply input the original data into the program, ignoring the fact that price limit may make the following day price data may be contaiminated.
73

Conversion de voix pour la synthèse de la parole

EN-NAJJARY, Taoufik 08 April 2005 (has links) (PDF)
Cette thèse s'inscrit dans le cadre des travaux de recherche entrepris par la division R&D de France Telecom dans le domaine de la synthèse de la parole à partir du texte. Elle concerne plus particulièrement le domaine de la conversion de voix, technologie visant à transformer le signal de parole d'un locuteur de référence dit locuteur source, de telle façon qu'il semble, à l'écoute, avoir été prononcé par un autre locuteur cible, identifié au préalable, dit locuteur cible. Le but de cette thèse est donc la diversification de voix de synthèse via la conception et le développement d'un système de conversion de voix de haute qualité. Les approches étudiées dans cette thèse se basent sur des techniques de classification par GMM (Gaussian Mixture Model) et une modélisation du signal de parole par HNM (Harmonic plus Noise Model). Dans un premier temps, l'influence de la paramétrisation spectrale sur la performance de conversion de voix par GMM est analysée. Puis, la dépendance entre l'enveloppe spectrale et la fréquence fondamentale est mise en évidence. Deux méthodes de conversion exploitant cette dépendance sont alors proposées et évaluées favorablement par rapport à l'état de l'art existant. Les problèmes liés à la mise en oeuvre de la conversion de voix sont également abordés. Le premier problème est la complexité élevée du processus de conversion par rapport au processus de synthèse lui-même (entre 1,5 et 2 fois le coût de calcul de la synthèse elle-même). Pour cela, une technique de conversion a été développée et conduit à une réduction de la complexité d'un facteur compris entre 45 et 130. Le deuxième problème concerne la mise en oeuvre de la conversion de voix lorsque les corpus d'apprentissage source et cible sont différents. Une méthodologie a ainsi été proposée rendant possible l'apprentissage de la fonction de transformation à partir d'enregistrements quelconques.
74

Quantization for Low Delay and Packet Loss

Subasingha, Subasingha Shaminda 22 April 2010 (has links)
Quantization of multimodal vector data in Realtime Interactive Communication Networks (RICNs) associated with application areas such as speech, video, audio, and haptic signals introduces a set of unique challenges. In particular, achieving the necessary distortion performance with minimum rate while maintaining low end-to-end delay and handling packet losses is of paramount importance. This dissertation presents vector quantization schemes which aim to satisfy these important requirements based on two source coding paradigms; 1) Predictive coding 2) Distributed source coding. Gaussian Mixture Models (GMMs) can be used to model any probability density function (pdf) with an arbitrarily small error given a sufficient number of mixture components. Hence, Gaussian Mixture Models can be effectively used to model the underlying pdfs of a variety of data in RICN applications. In this dissertation, first we present Gaussian Mixture Models Kalman predictive coding, which uses transform domain predictive GMM quantization techniques with Kalman filtering principles. In particular, we show how suitable modeling of quantization noise leads to a signal-adaptive GMM Kalman predictive coder that provides improved coding performance. Moreover, we demonstrate how running a GMM Kalman predictive coder to convergence can be used to design a stationary GMM Kalman predictive coding system which provides improved coding of GMM vector data but now with only a modest increase in run-time complexity over the baseline. Next, we address the issues of packet loss in the networks using GMM Kalman predictive coding principles. In particular, we show how an initial GMM Kalman predictive coder can be utilized to obtain a robust GMM predictive coder specifically designed to operate in packet loss. We demonstrate how one can define sets of encoding and decoding modes, and design special Kalman encoding and decoding gains for each mode. With this framework, GMM predictive coding design can be viewed as determining the special Kalman gains that minimize the expected mean squared error at the decoder in packet loss conditions. Finally, we present analytical techniques for modeling, analyzing and designing Wyner-Ziv(WZ) quantizers for Distributed Source Coding for jointly Gaussian vector data with imperfect side information. In most of the DSC implementations, the side information is not explicitly available in the decoder. Thus, almost all of the practical implementations obtain the side information from the previously decoded frames. Due to model imperfections, packet losses, previous decoding errors, and quantization noise, the available side information is usually noisy. However, the design of Wyner-Ziv quantizers for imperfect side information has not been widely addressed in the DSC literature. The analytical techniques presented in this dissertation explicitly assume the existence of imperfect side information in the decoder. Furthermore, we demonstrate how the design problem for vector data can be decomposed into independent scalar design subproblems. Then, we present the analytical techniques to compute the optimum step size and bit allocation for each scalar quantizer such that the decoder's expected vector Mean Squared Error(MSE) is minimized. The simulation results verify that the predicted MSE based on the presented analytical techniques closely follow the simulation results.
75

Essays on random effects models and GARCH

Skoglund, Jimmy January 2001 (has links)
This thesis consists of four essays, three in the field of random effects models and one in the field of GARCH. The first essay in this thesis, ''Maximum likelihood based inference in the two-way random effects model with serially correlated time effects'', considers maximum likelihood estimation and inference in the two-way random effects model with serial correlation. We derive a straightforward maximum likelihood estimator when the time-specific component follow an AR(1) or MA(1) process. The estimator is also easily generalized to allow for arbitrary stationary and strictly invertible ARMA processes. In addition we consider the model selection problem and derive tests of the null hypothesis of no serial correlation as well as tests for discriminating between the AR(1) and MA(1) specifications. A Monte-Carlo experiment evaluates the finite-sample properties of the estimators, test-statistics and model selection procedures. The second essay, ''Asymptotic properties of the maximum likelihood estimator of random effects models with serial correlation'', considers the large sample behavior of the maximum likelihood estimator of random effects models with serial correlation in the form of AR(1) for the idiosyncratic or time-specific error component. Consistent estimation and asymptotic normality is established for a comprehensive specification which nests these models as well as all commonly used random effects models. The third essay, ''Specification and estimation of random effects models with serial correlation of general form'', is also concerned with maximum likelihood based inference in random effects models with serial correlation. Allowing for individual effects we introduce serial correlation of general form in the time effects as well as the idiosyncratic errors. A straightforward maximum likelihood estimator is derived and a coherent model selection strategy is suggested for determining the orders of serial correlation as well as the importance of time or individual effects. The methods are applied to the estimation of a production function using a sample of 72 Japanese chemical firms observed during 1968-1987. The fourth essay, entitled ''A simple efficient GMM estimator of GARCH models'', considers efficient GMM based estimation of GARCH models. Sufficient conditions for the estimator to be consistent and asymptotically normal are established for the GARCH(1,1) conditional variance process. In addition efficiency results are obtained for a GARCH(1,1) model where the conditional variance is allowed to enter the mean as well. That is, the GARCH(1,1)-M model. An application to the returns to the SP500 index illustrates. / <p>Diss. Stockholm : Handelshögskolan, 2001</p>
76

The Impact Of Technology Level And Structural Change Of Exports On The Dynamics Of International Competitiveness: A Sectoral Disaggregated Analysis Of Turkish Manufacturing Sector

Sahan, Fatih 01 September 2012 (has links) (PDF)
The major aim of this thesis is to analyze the impact of structural change of exports and technology level on the international competitiveness. In order to analyze international competitiveness, export market shares are used. The empirical analysis suggested in this thesis includes two steps. In the first step, constant market share analysis is conducted to understand the causes of changes in export market shares from one period to another and in the second step a difference generalized method of moments model is proposed for 44 manufacturing sectors, which are classified with respect to their technology intensities, over 2003- 2008 period. The results are highly sensitive to the technology intensity of sectors.
77

Model Based Speech Enhancement and Coding

Zhao, David Yuheng January 2007 (has links)
In mobile speech communication, adverse conditions, such as noisy acoustic environments and unreliable network connections, may severely degrade the intelligibility and natural- ness of the received speech quality, and increase the listening effort. This thesis focuses on countermeasures based on statistical signal processing techniques. The main body of the thesis consists of three research articles, targeting two specific problems: speech enhancement for noise reduction and flexible source coder design for unreliable networks. Papers A and B consider speech enhancement for noise reduction. New schemes based on an extension to the auto-regressive (AR) hidden Markov model (HMM) for speech and noise are proposed. Stochastic models for speech and noise gains (excitation variance from an AR model) are integrated into the HMM framework in order to improve the modeling of energy variation. The extended model is referred to as a stochastic-gain hidden Markov model (SG-HMM). The speech gain describes the energy variations of the speech phones, typically due to differences in pronunciation and/or different vocalizations of individual speakers. The noise gain improves the tracking of the time-varying energy of non-stationary noise, e.g., due to movement of the noise source. In Paper A, it is assumed that prior knowledge on the noise environment is available, so that a pre-trained noise model is used. In Paper B, the noise model is adaptive and the model parameters are estimated on-line from the noisy observations using a recursive estimation algorithm. Based on the speech and noise models, a novel Bayesian estimator of the clean speech is developed in Paper A, and an estimator of the noise power spectral density (PSD) in Paper B. It is demonstrated that the proposed schemes achieve more accurate models of speech and noise than traditional techniques, and as part of a speech enhancement system provide improved speech quality, particularly for non-stationary noise sources. In Paper C, a flexible entropy-constrained vector quantization scheme based on Gaus- sian mixture model (GMM), lattice quantization, and arithmetic coding is proposed. The method allows for changing the average rate in real-time, and facilitates adaptation to the currently available bandwidth of the network. A practical solution to the classical issue of indexing and entropy-coding the quantized code vectors is given. The proposed scheme has a computational complexity that is independent of rate, and quadratic with respect to vector dimension. Hence, the scheme can be applied to the quantization of source vectors in a high dimensional space. The theoretical performance of the scheme is analyzed under a high-rate assumption. It is shown that, at high rate, the scheme approaches the theoretically optimal performance, if the mixture components are located far apart. The practical performance of the scheme is confirmed through simulations on both synthetic and speech-derived source vectors. / QC 20100825
78

Essays on Value-Added Taxation

El-Ganainy, Asmaa Adel 08 August 2006 (has links)
This dissertation evaluates the empirical relation between the value-added tax (VAT) and the level of aggregate consumption. Furthermore, it develops a theoretical framework and an empirical analysis to study the impact of the VAT, as a form of taxing consumption, on capital accumulation, productivity growth, and overall economic growth. While recent theoretical work shows that the VAT may boost capital accumulation and growth by encouraging more savings, we find that the net impact of consumption taxes on growth and its sources is theoretically ambiguous, and depends on the interaction between utility parameters, the interest rate, and the tax structure. Moreover, we develop a theoretical model to study the tax design problem in order to rationalize the observed variation in effective VAT rates over time in our sample. This framework considers both equity and efficiency as important factors determining optimal tax structure, and we identify conditions under which taxes could be evolving or constant over time. Empirically, we use a panel of 15 European Union countries and employ the recently developed GMM dynamic panel techniques. After controlling for the potential biases associated with persistence, endogeneity, simultaneity, measurement error, omitted variables, and unobserved country-specific effects, we find that (i) the VAT exerts a negative impact on the level of aggregate consumption, (ii) the VAT affects physical capital accumulation positively, which feeds through to overall GDP growth, and (iii) productivity growth seems to be a less relevant channel for the VAT to influence economic growth.
79

Développement financier, instabilité financière et croissance économique : implications pour la réduction de la pauvreté

Kpodar, Kangni 23 October 2006 (has links) (PDF)
La littérature sur la relation entre le développement financier et la croissance est très vaste et ancienne. Selon la théorie économique, le développement de l'intermédiation est favorable à la croissance économique car l'activité des banques accroît la mobilisation de l'épargne, améliore l'efficacité de l'allocation des ressources, et stimule l'innovation technologique. Cependant, certaines expériences de politiques visant à libéraliser les systèmes financiers des contraintes qui les empêchent de se développer et de contribuer à la croissance, se sont soldées par des échecs, ce qui a conduit à jeter un certain doute sur la généralité de la relation entre développement financier et développement économique. Ce doute a persisté avec les nombreuses études appliquées. Si la plupart des études (y compris les plus récentes) ont pu mettre en lumière, conformément aux prédictions théoriques, une relation positive entre le développement financier et la croissance (par exemple, Levine, Loayza et Beck, 2000), d'autres études (Andersen et Tarp (2003) par exemple) ont suggéré que la relation entre le développement financier pourrait être moins générale que ne le pense la littérature traditionnelle et ont souligné notamment que les résultats des études économétriques varient en fonction de l'échantillon et de la période considérée.<br />Dans cette thèse, nous nous sommes intéressés aux raisons pouvant expliquer les résultats ambigus des études appliquées sur le lien entre le développement financier et la croissance. En premier lieu, nous considérons que le développement financier risque d'être simultanément une source d'instabilité financière de telle sorte que l'effet bénéfique du développement financier sur la croissance soit amoindri ; il nous paraît donc indispensable de prendre en compte ce lien entre le développement financier et l'instabilité financière pour pouvoir véritablement apprécier la contribution du développement financier à la croissance. En second lieu, nous considérons l'existence d'effets de seuil dans la relation entre le développement financier et la croissance. En effet, il se peut qu'il existe un seuil minimum de développement économique en dessous duquel le développement financier n'a pas d'impact significatif sur la croissance, principalement à cause de la faiblesse de l'épargne et de la rentabilité des investissements. Par ailleurs, étant donné que la littérature économique s'est beaucoup consacrée à la relation entre le développement financier et la croissance, et très peu à la relation entre le développement financier et la réduction de la pauvreté, nous nous sommes également intéressés à l'impact spécifique que le développement financier peut avoir sur la réduction de la pauvreté au-delà de son effet indirect qui passe par la croissance.<br />L'analyse économétrique effectuée sur un panel de pays en développement avec des données quinquennales sur la période 1966-2000 nous a permis de mettre en évidence une relation positive entre le niveau de développement financier et celui de l'instabilité financière ; en particulier, l'instabilité du niveau de développement financier et l'occurrence de crises bancaires s'accroissent avec le développement du système financier. Les résultats montrent également que l'instabilité financière a un effet négatif sur la croissance économique et qu'elle réduit l'impact favorable du développement financier sur la croissance sans toutefois l'annuler. Par ailleurs, il ressort de notre analyse que pour les pays dont le niveau de PIB par tête est inférieur à un seuil de 2560 dollars, le développement financier ne semble pas avoir d'impact significatif sur la croissance. L'existence des effets de seuil et la prise en compte de l'instabilité financière dans la relation entre le développement financier et la croissance constituent des hypothèses complémentaires permettant d'expliquer les résultats ambigus des études appliquées sur le lien entre le développement financier et la croissance. Enfin, notre analyse montre également qu'en plus de son effet à travers la croissance, le développement financier favorise la réduction de la pauvreté principalement grâce à l'effet de conduit du capital de McKinnon (1973), l'accès aux dépôts profitent plus aux pauvres que l'accès aux crédits.
80

Evaluation of Bone Contrast Enhanced MRI Sequences and Voxel Based Segmentation

Johansson, Adam January 2010 (has links)
An ultra-short echo time (UTE) magnetic resonance imaging (MRI) sequence was used together with other MRI sequences to evaluate the possibility of segmenting air, soft tissues and bone. Three patients were imaged with the UTE sequence and other sequences as well as with computed tomography (CT). An algorithm using Gaussian mixture models was developed and applied to the problem of segmenting the MR images. A similar algorithm was developed and used to generate an artificial CT image from the MR data. The images of the first patient were used as training data for the algorithms and the images of the other two patients were used for validation. It was found that less than 20 percent of the volume inside the head was misclassified and that the root mean square error of the artificial CT image was less than 420 Hounsfield units. Finally a volunteer was imaged in the same way but with an additional UTE sequence with a larger flip angle. The results suggested that the additional image may improve segmentation further. / En sekevens för bildgivande magnetresonans (MRI) med ultrakort ekotid (UTE) användes tillsammans med andra MRI-sekvenser till att utvärdera möjligheten att segmentera luft, mjukvävnad och ben. Bilder togs av tre patienter med UTE-sekvensen och med övriga sekvenser samt med datortomografi (CT). En algoritm baserad på en blanding av normalfördelningar utvecklades och tillämpades på MR-segmenteringsproblemet.En likande algoritm utvecklades och användes till att skapa en konstgjord CT-bild utifrån MR-bilderna.Bilderna tagna av den första patienten användes till att träna algoritmerna medan bilderna av de två andra patienterna användes för validering. Mindre än 20 procent av volymen inuti huvudet felklassificerades och det kvadratiska medelvärdet av avvikelserna i den konstgjorda CT-bilden var mindre än 420 hounsfieldenheter. Slutligen togs bilder av en frivillig på samma sätt men med ytterligare en UTE-sekvens med en större flippvinkel. Resultatet antyder att den nya bilden kan bidra till en förbättrad segmentering.

Page generated in 0.0981 seconds