Noise robust voice activity detection using features extracted from the time-domain autocorrelation function

This paper presents a method of voice activity detection (VAD) for high noise scenarios, using a noise robust voiced speech detection feature. The developed method is based on the fusion of two systems. The first system utilises the maximum peak of the normalised time-domain autocorrelation function (MaxPeak). The second zone system uses a novel combination of cross-correlation and zero-crossing rate of the normalised autocorrelation to approximate a measure of signal pitch and periodicity (CrossCorr) that is hypothesised to be noise robust. The score outputs by the two systems are then merged using weighted sum fusion to create the proposed autocorrelation zero-crossing rate (AZR) VAD. Accuracy of AZR was compared to state of the art and standardised VAD methods and was shown to outperform the best performing system with an average relative improvement of 24.8% in half-total error rate (HTER) on the QUT-NOISE-TIMIT database created using real recordings from high-noise environments.

Development of the response coefficient-root function method for the transient vibration serviceability design of flat concrete floors

The vibration serviceability limit state is an important design consideration for two-way, suspended concrete floors that is not always well understood by many practicing structural engineers. Although the field of floor vibration has been extensively developed, at present there are no convenient design tools that deal with this problem. Results from this research have enabled the development of a much-needed, new method for assessing the vibration serviceability of flat, suspended concrete floors in buildings. This new method has been named, the Response Coefficient-Root Function (RCRF) method. Full-scale, laboratory tests have been conducted on a post-tensioned floor specimen at Queensland University of Technology’s structural laboratory. Special support brackets were fabricated to perform as frictionless, pinned connections at the corners of the specimen. A series of static and dynamic tests were performed in the laboratory to obtain basic material and dynamic properties of the specimen. Finite-element-models have been calibrated against data collected from laboratory experiments. Computational finite-element-analysis has been extended to investigate a variety of floor configurations. Field measurements of floors in existing buildings are in good agreement with computational studies. Results from this parametric investigation have led to the development of new approach for predicting the design frequencies and accelerations of flat, concrete floor structures. The RCRF method is convenient tool to assist structural engineers in the design for the vibration serviceability limit-state of in-situ concrete floor systems.

Re-Presenting/Representing Melbourne: A literary and cultural analysis of Melbourne’s inner city and outer suburbs in contemporary fiction

This article provides an overview of a research project investigating contemporary literary representations of Melbourne’s inner and outer suburban spaces. It will argue that the city represented by local writers is an often more complex way of envisioning the city than the one presented in public policy and cultural discourses. In this view, the writer’s vision of a city does not necessarily override or provide a “truer’ account but it is in the fictional city where the complexity of the everyday life of a city is most accurately portrayed. The article will also provide an overview of the theoretical framework for reading the fictional texts in this way, examining how Soja’s concept of Thirdspace (2006) provides a place to engage “critically with theoretical issues, while simultaneously being that space where the debate occurs” (Mole 2008: 3).

Learning the kernel matrix with semidefinite programming

Kernel-based learning algorithms work by embedding the data into a Euclidean space, and then searching for linear relations among the embedded data points. The embedding is performed implicitly, by specifying the inner products between each pair of points in the embedding space. This information is contained in the so-called kernel matrix, a symmetric and positive semidefinite matrix that encodes the relative positions of all points. Specifying this matrix amounts to specifying the geometry of the embedding space and inducing a notion of similarity in the input space - classical model selection problems in machine learning. In this paper we show how the kernel matrix can be learned from data via semidefinite programming (SDP) techniques. When applied to a kernel matrix associated with both training and test data this gives a powerful transductive algorithm -using the labeled part of the data one can learn an embedding also for the unlabeled part. The similarity between test points is inferred from training points and their labels. Importantly, these learning problems are convex, so we obtain a method for learning both the model class and the function without local minima. Furthermore, this approach leads directly to a convex method for learning the 2-norm soft margin parameter in support vector machines, solving an important open problem.

Learning the Kernel Matrix with Semi-Definite Programming

Kernel-based learning algorithms work by embedding the data into a Euclidean space, and then searching for linear relations among the embedded data points. The embedding is performed implicitly, by specifying the inner products between each pair of points in the embedding space. This information is contained in the so-called kernel matrix, a symmetric and positive definite matrix that encodes the relative positions of all points. Specifying this matrix amounts to specifying the geometry of the embedding space and inducing a notion of similarity in the input space -- classical model selection problems in machine learning. In this paper we show how the kernel matrix can be learned from data via semi-definite programming (SDP) techniques. When applied to a kernel matrix associated with both training and test data this gives a powerful transductive algorithm -- using the labelled part of the data one can learn an embedding also for the unlabelled part. The similarity between test points is inferred from training points and their labels. Importantly, these learning problems are convex, so we obtain a method for learning both the model class and the function without local minima. Furthermore, this approach leads directly to a convex method to learn the 2-norm soft margin parameter in support vector machines, solving another important open problem. Finally, the novel approach presented in the paper is supported by positive empirical results.

Loss-of-function fibroblast growth factor receptor-2 mutations in melanoma

We report that 10% of melanoma tumors and cell lines harbor mutations in the fibroblast growth factor receptor 2 (FGFR2) gene. These novel mutations include three truncating mutations and 20 missense mutations occurring at evolutionary conserved residues in FGFR2 as well as among all four FGFRs. The mutation spectrum is characteristic of those induced by UV radiation. Mapping of these mutations onto the known crystal structures of FGFR2 followed by in vitro and in vivo studies show that these mutations result in receptor loss of function through several distinct mechanisms, including loss of ligand binding affinity, impaired receptor dimerization, destabilization of the extracellular domains, and reduced kinase activity. To our knowledge, this is the first demonstration of loss-of-function mutations in a class IV receptor tyrosine kinase in cancer. Taken into account with our recent discovery of activating FGFR2 mutations in endometrial cancer, we suggest that FGFR2 may join the list of genes that play context-dependent opposing roles in cancer.