This paper presents text-independent speaker identification results for medium-size speaker population sizes up to 400 speakers for TV speech and TIMIT database . A system based on Gaussian mixture speaker models is used for speaker identification, and experiments are conducted on the TV database and TIMIT database. The TV-Database results show medium-size population performance under TV conditions. These are believed to be the first speaker identification experiments on the complete 400 speaker TV databases and the largest text-independent speaker identification task reported to date. Identification accuracies of 94.5% on the TV databases, respectively and 98.5% on the TIMIT database .

A Mandarin keyword spotting system based on MFCC, discrete-time HMM and Viterbi algorithm with DTW is proposed in this thesis. Joining with a dialogue system, this keyword spotting platform is further refined to a prototype of natural speech patient registration system of Kaohsiung Veterans General Hospital. After the ID number is asked by the computer-dialogue attendant in the registration process, the user can finish all relevant works in one sentence. Functions of searching clinical doctors, making and canceling registration are all built in this system. In a laboratory environment, the correct rate of this speaker-independent patient registration system can reach 97% and all registration process can be completed within 75 seconds.

This thesis presents the dynamic and static forced performance of a flexure-pivot tilting-pad bearing load between pad (LBP) configuration for different rotor speeds and bearing unit loadings. The bearing has the following design parameters: 4 pads with pad arc angle 72o and 50% pivot offset, pad axial length 0.0762 m (3 in), pad radial clearance 0.254 mm (0.010 in), bearing radial clearance 0.1905 mm (0.0075 in), preload 0.25 and shaft nominal diameter of 0.11684 m (4.600 in). The dynamic coefficients and the static performance parameters of the FPB have been compared with the theoretical predictions using the isothermal analysis from the rotordynamic software suite XLTRC2-XLTFPBrg. The bearing shows a small attitude angle, about 10o, which indicates small crosscoupling stiffnesses. The pad temperatures increase in the circumferential direction of rotation with speed and load. The pads maximum temperature was measured near the trailing edge. The dependency of the stiffness and damping coefficients on the excitation frequency has been studied. The frequency dependency in the dynamic coefficients was removed by introducing an added mass coefficient to the bearing model. The direct added mass coefficients were around 32 kg. The direct stiffness and damping coefficients increase with load, while increasing and decreasing with rotor speed, respectively. A small whirl frequency ratio (WFR) was found of about 0.15, and it decreases with load and increases with speed. A comparison between the dynamic stiffnesses using a Reynolds equation and the bulk-flow Navier-Stokes models with the experimental dynamic stiffnesses shows that the Reynolds model (even for laminar flows) is not adequate, and that the bulk-flow model should be used for rotordynamic coefficients prediction. The bulk-flow model in general predicts well the static performance parameters and the direct dynamic coefficients, and underpredicts the cross-coupled coefficients (overpredicts the stability).

This thesis investigates the design and implementation strategies for a Taiwanese speech recognition system. It adopts a 4 plus 1¡]five times¡^recording strategy, where the 1st four recordings are used for speech feature training and the last recording for speech recognition simulation. Mel-frequency cepstrum coefficients and hidden Markov model are used as the feature model and the recognition model respectively. Under the Intel Celeron 2.4 GHz personal computer and Red Hat Linux 9.0 operating system environment, a correct phrase recognition rate of 90% can be reached for a 4200 Taiwanese phrase database.

One of the communities in the path of Hurricane Charley as it came ashore August 13, 2004, was Punta Gorda, recording gusts up to 145 mph. This project utilizes aerial photos taken approximately 10 days after the storm battered the area, using a digital photography program. Focusing on the one-story residential structures (houses) of the Punta Gorda area, a damage assessment could be made of the area’s homes, and how they stood up to the storm. This study focused further on homes built after major changes to the local/state building codes went into effect (starting in 1996) after the devastation left in south Florida by Hurricane Andrew in 1992. After selecting approximately 20 damaged houses, damaged from wind loading only, an analysis of these houses (or types of houses) could then be undertaken complying with the most current building/wind codes used at the time of Charley’s landfall. Furthermore, by looking at the pictures, and using reports outlining the types of damage seen from the storm, the building/wind codes could then be checked for their effectiveness. After performing a wind loading analysis on houses similar to those seen in the selected pictures, and using the wind code provisions of ASCE 7-98, calculations show a substantial increase in local wind pressure to various zones of the roof. High pressure zones of the roof included the ridges of the gable and hipped style roofs, as well as the corners and the edges. More emphasis needs to be placed on the installation of the clay tiles (mandated by certain deed-restricted subdivisions of Punta Gorda). If the tiles are ripped off from the wind, then the roof sheathing becomes exposed to the environment, and if this becomes damaged, rain leaking down into the interior of the house would cause additional damage.

In this dissertation we present four papers as chapters. In Chapter 2, we extended the techniques used for the Klein-Gordon Chain by Iooss, Kirchgässner, James, and Sire, to chains with non-nearest neighbor interactions. We look for travelling waves by reducing the Klein-Gordon chain with second nearest neighbor interaction to an advance-delay equation. Then we reduce the equation to a finite dimensional center manifold for some parameter regimes. By using the normal form expansion on the center manifold we were able to prove the existence of three different types of travelling solutions for the Klein Gordon Chain: periodic, quasi-periodic and homoclinic to periodic orbits with exponentially small amplitude. In Chapter 3 we include numerical methods for computing quasi-periodic solutions. We developed very efficient algorithms to compute smooth quasiperiodic equilibrium states of models in 1-D statistical mechanics models allowing non-nearest neighbor interactions. If we discretize a hull function using N Fourier coefficients, the algorithms require O(N) storage and a Newton step for the equilibrium equation requires only O(N log(N)) arithmetic operations. This numerical methods give rise to a criterion for the breakdown of quasi-periodic solutions. This criterion is presented in Chapter 4. In Chapter 5, we justify rigorously the criterion in Chapter 4. The justification of the criterion uses both Numerical KAM algorithms and rigorous results. The hypotheses of the theorem concern bounds on the Sobolev norms of a hull function and can be verified rigorously by the computer. The argument works with small modifications in all cases where there is an a posteriori KAM theorem. / text

Buvo parodyta, kaip gali būti naudojama JAHN programa darbui su atominėmis ir branduolinėmis būsenos funkcijomis izosukinio bazėje. Visa informacija apie atomines ar konfigūracines būsenos funkcijas gali būti pateikta įprastame formate. Tokia tvarka leistų greitą priėjimą prie reikiamų duomenų ir paprastą komandų valdymą. Būsenos funkcijų klasifikacijas izosukinio bazėje, viendalelius kilminius koeficientus CFP, dvidalelius kilminius koeficientus CFGP, transformacijos matricas iš LS ryšio į izosukinio ryšio bazę galima surasti naudojant atitinkamas komandas, kurių sintaksė yra analogiška literatūroje naudojamiems žymėjimams. Trumpas ir efektyvus sluoksnių ir konfigūracinių būsenų įvedimas daro programą atsparesnę klaidoms ir leidžia vartotojui sukurti atskiras koeficientų lenteles tam tikriems sluoksnių būsenų termams izosukinio bazėje. Tokios lentelės gaunamos vienos komandinės eilutės pagalba. Tikimasi, kad JAHN programa padės smulkiau analizuoti įvairių fizikinių operatorių matricinius elementus izosukinio bazėje. Komandos jj ryšiui, analogiškai kaip ir LS ryšiui, yra pateikiamos kuo mažiau nutolstant nuo jų užrašymo formos knygose ar straipsniuose. Įvedant nesudėtingas komandas nustatome sluoksnio charakteristikas. CFGP reikšmės irgi randamos JAHN programa, kas gerokai sutaupo laiko, kadangi jų nereikia ieškoti spausdinamose lentelėse. / We have shown how JAHN has been used for dealing with atomic or nuclear state functions within an isospin basis. All information about the atomic or the configuration state functions can be stored in a suitable format in order to allow a fast access and manipulation. Apart from the classification of the shell states in an isospin basis the coefficients of fractional parentage, the coefficients of fractional grandparentage, the transformation matrices for going from LS – coupled to an isospin – coupled basis can be calculated interactively. A short and powerful notation has been introduced for the shell and configuration states which makes the program less prone to errors and which enables the user to create an electronic table of shell state terms or CFP in isospin basis within only a single command line. The JAHN program will support a more detailed analysis of the matrix elements of various physical operators by using an isospin basis. The algorithms for jj coupling is analogous with LS coupling and they are presented in the way, which is commonly used in printed books and articles. We define layer characteristics from ordinary commands. CFGP values are also calculated with JAHN program, which is very convenient for saving the time that is needed to look up for those values in the printed tables.

This thesis deals with the numerical solution of ordinary differential equations (ODEs) using finite difference (FD) methods. In particular, boundary summation equation (BSE) preconditioning for FD approximations for ODEs with constant coefficients on locally refined meshes is studied. Firstly, the BSE for FD approximations of ODEs with constant coefficients is derived on a locally refined mesh. Secondly, the obtained linear system of equations are solved by the iterative method GMRES. Then, the arithmetic complexity and convergence rate of the iterative solution of the BSE formulation are discussed. Finally, numerical experiments are performed to compare the new approach with the FD approach. The results show that the BSE formulation has low arithmetic complexity and the convergence rate of the iterative solvers is fast and independent of the number of grid points.

In the modern day, recognition and tracking of face or the iris is potentially one of the most powerful ways of differentiating between an authentic person and an imposter. Our method uses stereo vision to track the 3-Dimensional coordinates of a target equivalent to a person’s eyes and using a pan-tilt unit we target these areas for additional processing such as iris or facial imaging. One of the most important parts involved in tracking is the way the pan-tilt unit is calibrated. There have been techniques in the past where PTZ (Pan-tilt-zoom) digital camera has been used and calibrated using self calibration techniques involving a checker board calibration grid but the tracking error was found to be large in these techniques. We introduce a more accurate form of calibration of the pantilt unit using photogrammetric calibration technique and view the pan-tilt unit as an emulation of a Pinhole Lens Model to detect and track the target. The system is demonstrated on ideal targets.

Genitourinary cancer refers to the presence of tumours in the genital or urinary organs such as bladder, kidney and prostate. In 2008 the worldwide incidence of bladder cancer was 382,600 with a mortality of 150,282. Radiotherapy is one of the main treatment choices for genitourinary cancer where accurate delineation of the gross tumour volume (GTV) on computed tomography (CT) images is crucial for the success of this treatment. Limited CT resolution and contrast in soft tissue organs make this difficult and has led to significant inter- and intra- clinical variability in defining the extent of the GTV, especially at the junctions of different organs. In addition the introduction of new imaging techniques and modalities has significantly increased the number of the medical images that require contouring. More advanced image processing is required to help reduce contouring variability and assist in handling the increased volume of data. In this thesis image analysis methodologies were used to extract low-level features such as entropy, moment and correlation from radiotherapy planning CT images. These distinctive features were identified and used for defining the GTV and to implement a fully-automatic contouring system. The first key contribution is to demonstrate that second-order statistics from co-occurrence matrices (GTSDM) give higher accuracy in classifying soft tissue regions of interest (ROIs) into GTV and non-GTV. Loadings of the principal components (PCs) of the GTSDM features were found to be consistent over different patients. Exhaustive feature selection suggested that entropies and correlations produced consistently larger areas under receiver operating characteristic (AUROC) curves than first-order features. The second significant contribution is to demonstrate that in the bladder-prostate junction, where the largest inter-clinical variability is observed, the second-order principal entropy from stationery wavelet denoised CT images (DPE) increased the saliency of the bladder prostate junction. As a result thresholding of the DPE produced good agreement between gold standard clinical contours and those produced by this approach with Dice coefficients. The third contribution is to implement a fully automatic and reproducible system for bladder cancer GTV auto-contouring based on classifying second-order statistics. The Dice similarity coefficients (DSCs) were employed to evaluate the automatic contours. It was found that in the mid-range of the bladder the automatic contours are accurate, but in the inferior and superior ends of bladder automatic contours were more likely to have small DSCs with clinical contours, which reconcile with the fact of clinical variability in defining GTVs. A novel male bladder probability atlas was constructed based on the clinical contours and volume estimation from the classification results. Registration of the classification results with this probabilistic atlas consistently increases the DSCs of the inferior slices.