Improved GMM-based speaker verification using SVM-driven impostor dataset selection

The problem of impostor dataset selection for GMM-based speaker verification is addressed through the recently proposed data-driven background dataset refinement technique. The SVM-based refinement technique selects from a candidate impostor dataset those examples that are most frequently selected as support vectors when training a set of SVMs on a development corpus. This study demonstrates the versatility of dataset refinement in the task of selecting suitable impostor datasets for use in GMM-based speaker verification. The use of refined Z- and T-norm datasets provided performance gains of 15% in EER in the NIST 2006 SRE over the use of heuristically selected datasets. The refined datasets were shown to generalise well to the unseen data of the NIST 2008 SRE.

Data-driven impostor selection for T-norm score normalisation and the background dataset in SVM-based speaker verification

A data-driven background dataset refinement technique was recently proposed for SVM based speaker verification. This method selects a refined SVM background dataset from a set of candidate impostor examples after individually ranking examples by their relevance. This paper extends this technique to the refinement of the T-norm dataset for SVM-based speaker verification. The independent refinement of the background and T-norm datasets provides a means of investigating the sensitivity of SVM-based speaker verification performance to the selection of each of these datasets. Using refined datasets provided improvements of 13% in min. DCF and 9% in EER over the full set of impostor examples on the 2006 SRE corpus with the majority of these gains due to refinement of the T-norm dataset. Similar trends were observed for the unseen data of the NIST 2008 SRE.

Robust video stabilisation algorithm using feature point selection and delta optical flow

In this study, the authors propose a novel video stabilisation algorithm for mobile platforms with moving objects in the scene. The quality of videos obtained from mobile platforms, such as unmanned airborne vehicles, suffers from jitter caused by several factors. In order to remove this undesired jitter, the accurate estimation of global motion is essential. However it is difficult to estimate global motions accurately from mobile platforms due to increased estimation errors and noises. Additionally, large moving objects in the video scenes contribute to the estimation errors. Currently, only very few motion estimation algorithms have been developed for video scenes collected from mobile platforms, and this paper shows that these algorithms fail when there are large moving objects in the scene. In this study, a theoretical proof is provided which demonstrates that the use of delta optical flow can improve the robustness of video stabilisation in the presence of large moving objects in the scene. The authors also propose to use sorted arrays of local motions and the selection of feature points to separate outliers from inliers. The proposed algorithm is tested over six video sequences, collected from one fixed platform, four mobile platforms and one synthetic video, of which three contain large moving objects. Experiments show our proposed algorithm performs well to all these video sequences.

The applicability of biased estimation in model and model order selection

Biased estimation has the advantage of reducing the mean squared error (MSE) of an estimator. The question of interest is how biased estimation affects model selection. In this paper, we introduce biased estimation to a range of model selection criteria. Specifically, we analyze the performance of the minimum description length (MDL) criterion based on biased and unbiased estimation and compare it against modern model selection criteria such as Kay's conditional model order estimator (CME), the bootstrap and the more recently proposed hook-and-loop resampling based model selection. The advantages and limitations of the considered techniques are discussed. The results indicate that, in some cases, biased estimators can slightly improve the selection of the correct model. We also give an example for which the CME with an unbiased estimator fails, but could regain its power when a biased estimator is used.

Performing composition : developing a computer assisted composition system through live coding performance

This thesis maps the author's journey from a music composition practice to a composition and performance practice. The work involves the development of a software library for the purpose of encapsulating compositional ideas in software, and realising these ideas in performance through a live coding computer music practice. The thesis examines what artistic practice emerges through live coding and software development, and does this permit a blurring between the activities of music composition and performance. The role that software design plays in affecting musical outcomes is considered to gain an insight into how software development contributes to artistic development. The relationship between music composition and performance is also examined to identify the means by which engaging in live coding and software development can bring these activities together. The thesis, situated within the discourse of practice led research, documents a journey which uses the experience of software development and performance as a means to guide the direction of the research. The journey serves as an experiment for the author in engaging an hitherto unfamiliar musical practice, and as a roadmap for others seeking to modify or broaden their artistic practice.

Data-driven background dataset selection for SVM-based speaker verification

The recently proposed data-driven background dataset refinement technique provides a means of selecting an informative background for support vector machine (SVM)-based speaker verification systems. This paper investigates the characteristics of the impostor examples in such highly-informative background datasets. Data-driven dataset refinement individually evaluates the suitability of candidate impostor examples for the SVM background prior to selecting the highest-ranking examples as a refined background dataset. Further, the characteristics of the refined dataset were analysed to investigate the desired traits of an informative SVM background. The most informative examples of the refined dataset were found to consist of large amounts of active speech and distinctive language characteristics. The data-driven refinement technique was shown to filter the set of candidate impostor examples to produce a more disperse representation of the impostor population in the SVM kernel space, thereby reducing the number of redundant and less-informative examples in the background dataset. Furthermore, data-driven refinement was shown to provide performance gains when applied to the difficult task of refining a small candidate dataset that was mis-matched to the evaluation conditions.

Exploiting multiple feature sets in data-driven impostor dataset selection for speaker verification

This study assesses the recently proposed data-driven background dataset refinement technique for speaker verification using alternate SVM feature sets to the GMM supervector features for which it was originally designed. The performance improvements brought about in each trialled SVM configuration demonstrate the versatility of background dataset refinement. This work also extends on the originally proposed technique to exploit support vector coefficients as an impostor suitability metric in the data-driven selection process. Using support vector coefficients improved the performance of the refined datasets in the evaluation of unseen data. Further, attempts are made to exploit the differences in impostor example suitability measures from varying features spaces to provide added robustness.

The home stigma : adverse selection in ADRs and the home capital market environment

We investigate whether characteristics of the home country capital environment, such as information disclosure and investor rights protection continue to affect ADRs cross-listed in the U.S. Using microstructure measures as proxies for adverse selection, we find that characteristics of the home markets continue to be relevant, especially for emerging market firms. Less transparent disclosure, poorer protection of investor rights and weaker legal institutions are associated with higher levels of information asymmetry. Developed market firms appear to be affected by whether or not home business laws are common law or civil law legal origin. Our finding contributes to the bonding literature. It suggests that cross-listing in the U.S. should not be viewed as a substitute for improvement in the quality of local institutions, and attention must be paid to improve investor protection in order to achieve the full benefits of improved disclosure. Improvement in the domestic capital market environment can attract more investors even for U.S. cross-listed firms.

Secondary curve behaviour in Lenke IC class adolescent idiopathic scoliosis following video assisted thoracoscopic spinal fusion

Introduction. Ideally after selective thoracic fusion for Lenke Class IC (i.e. major thoracic / secondary lumbar) curves, the lumbar spine will spontaneously accommodate to the corrected position of the thoracic curve, thereby achieving a balanced spine, avoiding the need for fusion of lumbar spinal segments1. The purpose of this study was to evaluate the behaviour of the lumbar curve in Lenke IC class adolescent idiopathic scoliosis (AIS) following video-assisted thoracoscopic spinal fusion and instrumentation (VATS) of the major thoracic curve. Methods. A retrospective review of 22 consecutive patients with AIS who underwent VATS by a single surgeon was conducted. The results were compared to published literature examining the behaviour of the secondary lumbar curve where other surgical approaches were employed. Results. Twenty-two patients (all female) with AIS underwent VATS. All major thoracic curves were right convex. The average age at surgery was 14 years (range 10 to 22 years). On average 6.7 levels (6 to 8) were instrumented. The mean follow-up was 25.1 months (6 to 36). The pre-operative major thoracic Cobb angle mean was 53.8° (40° to 75°). The pre-operative secondary lumbar Cobb angle mean was 43.9° (34° to 55°). On bending radiographs, the secondary curve corrected to 11.3° (0° to 35°). The rib hump mean measurement was 15.0° (7° to 21°). At latest follow-up the major thoracic Cobb angle measured on average 27.2° (20° to 41°) (p<0.001 – univariate ANOVA) and the mean secondary lumbar curve was 27.3° (15° to 42°) (p<0.001). This represented an uninstrumented secondary curve correction factor of 37.8%. The mean rib hump measured was 6.5° (2° to 15°) (p<0.001). The results above were comparable to published series when open surgery was performed. Discussion. VATS is an effective method of correcting major thoracic curves with secondary lumbar curves. The behaviour of the secondary lumbar curve is consistent with published series when open surgery, both anterior and posterior, is performed.

Computed tomography analysis of sagittal plane parameters in adolescent idiopathic scoliosis treated by video assisted thoracoscopic spinal fusion and instrumentation

One of the primary treatment goals of adolescent idiopathic scoliosis (AIS) surgery is to achieve maximum coronal plane correction while maintaining coronal balance. However maintaining or restoring sagittal plane spinal curvature has become increasingly important in maintaining the long-term health of the spine. Patients with AIS are characterised by pre-operative thoracic hypokyphosis, and it is generally agreed that operative treatment of thoracic idiopathic scoliosis should aim to restore thoracic kyphosis to normal values while maintaining lumbar lordosis and good overall sagittal balance. The aim of this study was to evaluate CT sagittal plane parameters, with particular emphasis on thoracolumbar junctional alignment, in patients with AIS who underwent Video Assisted Thoracoscopic Spinal Fusion and Instrumentation (VATS). This study concluded that video-assisted thoracoscopic spinal fusion and instrumentation reliably increases thoracic kyphosis while preserving junctional alignment and lumbar lordosis in thoracic AIS.

Peer assisted study sessions

My journey with Peer Assisted Study Sessions, or Supplemental Instruction (SI), began in 1993 when I took over a 1st year, 1st semester unit in QUT's Bachelor of Engineering program. The unit had 500 enrolments with students from all 10 engineering majors at QUT. The 500 students received a 2 hour lecture and a 1 hour tutorial per week, usually run by academic staff or postgraduate students. The unit covered basic mechanics, which comprises a challenging set of topics on how forces interact with various bodies. One normally expects 1st year students to find it difficult to come to grips with the material. However, when I ran that unit in 1993, the failure rate had been usually around 50%.