QUT Speaker Identity Verification system for EVALITA 2009

This document outlines the system submitted by the Speech and Audio Research Laboratory at the Queensland University of Technology (QUT) for the Speaker Identity Verification: Application task of EVALITA 2009. This competitive submission consisted of a score-level fusion of three component systems; a joint-factor analysis GMM system and two SVM systems using GLDS and GMM supervector kernels. Development evaluation and post-submission results are presented in this study, demonstrating the effectiveness of this fused system approach. This study highlights the challenges associated with system calibration from limited development data and that mismatch between training and testing conditions continues to be a major source of error in speaker verification technology.

Machine learning techniques to improve software quality

A significant proportion of the cost of software development is due to software testing and maintenance. This is in part the result of the inevitable imperfections due to human error, lack of quality during the design and coding of software, and the increasing need to reduce faults to improve customer satisfaction in a competitive marketplace. Given the cost and importance of removing errors improvements in fault detection and removal can be of significant benefit. The earlier in the development process faults can be found, the less it costs to correct them and the less likely other faults are to develop. This research aims to make the testing process more efficient and effective by identifying those software modules most likely to contain faults, allowing testing efforts to be carefully targeted. This is done with the use of machine learning algorithms which use examples of fault prone and not fault prone modules to develop predictive models of quality. In order to learn the numerical mapping between module and classification, a module is represented in terms of software metrics. A difficulty in this sort of problem is sourcing software engineering data of adequate quality. In this work, data is obtained from two sources, the NASA Metrics Data Program, and the open source Eclipse project. Feature selection before learning is applied, and in this area a number of different feature selection methods are applied to find which work best. Two machine learning algorithms are applied to the data - Naive Bayes and the Support Vector Machine - and predictive results are compared to those of previous efforts and found to be superior on selected data sets and comparable on others. In addition, a new classification method is proposed, Rank Sum, in which a ranking abstraction is laid over bin densities for each class, and a classification is determined based on the sum of ranks over features. A novel extension of this method is also described based on an observed polarising of points by class when rank sum is applied to training data to convert it into 2D rank sum space. SVM is applied to this transformed data to produce models the parameters of which can be set according to trade-off curves to obtain a particular performance trade-off.

A new unequal DC link voltage configuration for a single phase multilevel converter to reduce low order harmonics

Multilevel converters are used in high power and high voltage applications due to their attractive benefits in generating high quality output voltage. Increasing the number of voltage levels can lead to a reduction in lower order harmonics. Various modulation and control techniques are introduced for multilevel converters like Space Vector Modulation (SVM), Sinusoidal Pulse Width Modulation (SPWM) and Harmonic Elimination (HE) methods. Multilevel converters may have a DC link with equal or unequal DC voltages. In this paper a new modulation technique based on harmonic elimination method is proposed for those multilevel converters that have unequal DC link voltages. This new technique has better effect on output voltage quality and less Total Harmonic Distortion (THD) than other modulation techniques. In order to verify the proposed modulation technique, MATLAB simulations are carried out for a single-phase diode-clamped inverter.

Real-time evaluation of driver’s alertness on highways

Suburbanisation has been internationally a major phenomenon in the last decades. Suburb-to-suburb routes are nowadays the most widespread road journeys; and this resulted in an increment of distances travelled, particularly on faster suburban highways. The design of highways tends to over-simplify the driving task and this can result in decreased alertness. Driving behaviour is consequently impaired and drivers are then more likely to be involved in road crashes. This is particularly dangerous on highways where the speed limit is high. While effective countermeasures to this decrement in alertness do not currently exist, the development of in-vehicle sensors opens avenues for monitoring driving behaviour in real-time. The aim of this study is to evaluate in real-time the level of alertness of the driver through surrogate measures that can be collected from in-vehicle sensors. Slow EEG activity is used as a reference to evaluate driver's alertness. Data are collected in a driving simulator instrumented with an eye tracking system, a heart rate monitor and an electrodermal activity device (N=25 participants). Four different types of highways (driving scenario of 40 minutes each) are implemented through the variation of the road design (amount of curves and hills) and the roadside environment (amount of buildings and traffic). We show with Neural Networks that reduced alertness can be detected in real-time with an accuracy of 92% using lane positioning, steering wheel movement, head rotation, blink frequency, heart rate variability and skin conductance level. Such results show that it is possible to assess driver's alertness with surrogate measures. Such methodology could be used to warn drivers of their alertness level through the development of an in-vehicle device monitoring in real-time drivers' behaviour on highways, and therefore it could result in improved road safety.

Driving performance impairments due to hypovigilance on monotonous roads

Drivers' ability to react to unpredictable events deteriorates when exposed to highly predictable and uneventful driving tasks. Highway design reduces the driving task mainly to a lane-keeping manoeuvre. Such a task is monotonous, providing little stimulation and this contributes to crashes due to inattention. Research has shown that driver's hypovigilance can be assessed with EEG measurements and that driving performance is impaired during prolonged monotonous driving tasks. This paper aims to show that two dimensions of monotony - namely road design and road side variability - decrease vigilance and impair driving performance. This is the first study correlating hypovigilance and driver performance in varied monotonous conditions, particularly on a short time scale (a few seconds). We induced vigilance decrement as assessed with an EEG during a monotonous driving simulator experiment. Road monotony was varied through both road design and road side variability. The driver's decrease in vigilance occurred due to both road design and road scenery monotony and almost independently of the driver's sensation seeking level. Such impairment was also correlated to observable measurements from the driver, the car and the environment. During periods of hypovigilance, the driving performance impairment affected lane positioning, time to lane crossing, blink frequency, heart rate variability and non-specific electrodermal response rates. This work lays the foundation for the development of an in-vehicle device preventing hypovigilance crashes on monotonous roads.

Cardiac health diagnosis using higher order spectra and support vector machine

The Electrocardiogram (ECG) is an important bio-signal representing the sum total of millions of cardiac cell depolarization potentials. It contains important insight into the state of health and nature of the disease afflicting the heart. Heart rate variability (HRV) refers to the regulation of the sinoatrial node, the natural pacemaker of the heart by the sympathetic and parasympathetic branches of the autonomic nervous system. The HRV signal can be used as a base signal to observe the heart's functioning. These signals are non-linear and non-stationary in nature. So, higher order spectral (HOS) analysis, which is more suitable for non-linear systems and is robust to noise, was used. An automated intelligent system for the identification of cardiac health is very useful in healthcare technology. In this work, we have extracted seven features from the heart rate signals using HOS and fed them to a support vector machine (SVM) for classification. Our performance evaluation protocol uses 330 subjects consisting of five different kinds of cardiac disease conditions. We demonstrate a sensitivity of 90% for the classifier with a specificity of 87.93%. Our system is ready to run on larger data sets.

Spectral–texture feature extraction using statistical moments with application to object-based vegetation species classification

The use of appropriate features to characterise an output class or object is critical for all classification problems. In order to find optimal feature descriptors for vegetation species classification in a power line corridor monitoring application, this article evaluates the capability of several spectral and texture features. A new idea of spectral–texture feature descriptor is proposed by incorporating spectral vegetation indices in statistical moment features. The proposed method is evaluated against several classic texture feature descriptors. Object-based classification method is used and a support vector machine is employed as the benchmark classifier. Individual tree crowns are first detected and segmented from aerial images and different feature vectors are extracted to represent each tree crown. The experimental results showed that the proposed spectral moment features outperform or can at least compare with the state-of-the-art texture descriptors in terms of classification accuracy. A comprehensive quantitative evaluation using receiver operating characteristic space analysis further demonstrates the strength of the proposed feature descriptors.

Comment

The support vector machine (SVM) has played an important role in bringing certain themes to the fore in computationally oriented statistics. However, it is important to place the SVM in context as but one member of a class of closely related algorithms for nonlinear classification. As we discuss, several of the “open problems” identified by the authors have in fact been the subject of a significant literature, a literature that may have been missed because it has been aimed not only at the SVM but at a broader family of algorithms. Keeping the broader class of algorithms in mind also helps to make clear that the SVM involves certain specific algorithmic choices, some of which have favorable consequences and others of which have unfavorable consequences—both in theory and in practice. The broader context helps to clarify the ties of the SVM to the surrounding statistical literature.

Exponentiated gradient algorithms for conditional random fields and max-margin Markov Networks

Log-linear and maximum-margin models are two commonly-used methods in supervised machine learning, and are frequently used in structured prediction problems. Efficient learning of parameters in these models is therefore an important problem, and becomes a key factor when learning from very large data sets. This paper describes exponentiated gradient (EG) algorithms for training such models, where EG updates are applied to the convex dual of either the log-linear or max-margin objective function; the dual in both the log-linear and max-margin cases corresponds to minimizing a convex function with simplex constraints. We study both batch and online variants of the algorithm, and provide rates of convergence for both cases. In the max-margin case, O(1/ε) EG updates are required to reach a given accuracy ε in the dual; in contrast, for log-linear models only O(log(1/ε)) updates are required. For both the max-margin and log-linear cases, our bounds suggest that the online EG algorithm requires a factor of n less computation to reach a desired accuracy than the batch EG algorithm, where n is the number of training examples. Our experiments confirm that the online algorithms are much faster than the batch algorithms in practice. We describe how the EG updates factor in a convenient way for structured prediction problems, allowing the algorithms to be efficiently applied to problems such as sequence learning or natural language parsing. We perform extensive evaluation of the algorithms, comparing them to L-BFGS and stochastic gradient descent for log-linear models, and to SVM-Struct for max-margin models. The algorithms are applied to a multi-class problem as well as to a more complex large-scale parsing task. In all these settings, the EG algorithms presented here outperform the other methods.

Prediction of cis/trans isomerization in proteins using PSI-BLAST profiles and secondary structure information

Background The majority of peptide bonds in proteins are found to occur in the trans conformation. However, for proline residues, a considerable fraction of Prolyl peptide bonds adopt the cis form. Proline cis/trans isomerization is known to play a critical role in protein folding, splicing, cell signaling and transmembrane active transport. Accurate prediction of proline cis/trans isomerization in proteins would have many important applications towards the understanding of protein structure and function. Results In this paper, we propose a new approach to predict the proline cis/trans isomerization in proteins using support vector machine (SVM). The preliminary results indicated that using Radial Basis Function (RBF) kernels could lead to better prediction performance than that of polynomial and linear kernel functions. We used single sequence information of different local window sizes, amino acid compositions of different local sequences, multiple sequence alignment obtained from PSI-BLAST and the secondary structure information predicted by PSIPRED. We explored these different sequence encoding schemes in order to investigate their effects on the prediction performance. The training and testing of this approach was performed on a newly enlarged dataset of 2424 non-homologous proteins determined by X-Ray diffraction method using 5-fold cross-validation. Selecting the window size 11 provided the best performance for determining the proline cis/trans isomerization based on the single amino acid sequence. It was found that using multiple sequence alignments in the form of PSI-BLAST profiles could significantly improve the prediction performance, the prediction accuracy increased from 62.8% with single sequence to 69.8% and Matthews Correlation Coefficient (MCC) improved from 0.26 with single local sequence to 0.40. Furthermore, if coupled with the predicted secondary structure information by PSIPRED, our method yielded a prediction accuracy of 71.5% and MCC of 0.43, 9% and 0.17 higher than the accuracy achieved based on the singe sequence information, respectively. Conclusion A new method has been developed to predict the proline cis/trans isomerization in proteins based on support vector machine, which used the single amino acid sequence with different local window sizes, the amino acid compositions of local sequence flanking centered proline residues, the position-specific scoring matrices (PSSMs) extracted by PSI-BLAST and the predicted secondary structures generated by PSIPRED. The successful application of SVM approach in this study reinforced that SVM is a powerful tool in predicting proline cis/trans isomerization in proteins and biological sequence analysis.

Improved sensorimotor adaptation after exhaustive exercise is accompanied by altered brain activity

Acute exercise has been shown to exhibit different effects on human sensorimotor behavior; however, the causes and mechanisms of the responses are often not clear. The primary aim of the present study was to determine the effects of incremental running until exhaustion on sensorimotor performance and adaptation in a tracking task. Subjects were randomly assigned to a running group (RG), a tracking group (TG), or a running followed by tracking group (RTG), with 10 subjects assigned to each group. Treadmill running velocity was initially set at 2.0 m s− 1, increasing by 0.5 m s− 1 every 5 min until exhaustion. Tracking consisted of 35 episodes (each 40 s) where the subjects' task was to track a visual target on a computer screen while the visual feedback was veridical (performance) or left-right reversed (adaptation). Resting electroencephalographic (EEG) activity was recorded before and after each experimental condition (running, tracking, rest). Tracking performance and the final amount of adaptation did not differ between groups. However, task adaptation was significantly faster in RTG compared to TG. In addition, increased alpha and beta power were observed following tracking in TG but not RTG although exhaustive running failed to induce significant changes in these frequency bands. Our results suggest that exhaustive running can facilitate adaptation processes in a manual tracking task. Attenuated cortical activation following tracking in the exercise condition was interpreted to indicate cortical efficiency and exercise-induced facilitation of selective central processes during actual task demands.

Electro-cortical implicit race bias does not vary with participants' race or sex

Earlier research found evidence for electro-cortical race bias towards black target faces in white American participants irrespective of the task relevance of race. The present study investigated whether an implicit race bias generalizes across cultural contexts and racial in- and out-groups. An Australian sample of 56 Chinese and Caucasian males and females completed four oddball tasks that required sex judgements for pictures of male and female Chinese and Caucasian posers. The nature of the background (across task) and of the deviant stimuli (within task) was fully counterbalanced. Event-related potentials (ERPs) to deviant stimuli recorded from three midline sites were quantified in terms of mean amplitude for four components: N1, P2, N2 and a late positive complex (LPC; 350–700 ms). Deviants that differed from the backgrounds in sex or race elicited enhanced LPC activity. These differences were not modulated by participant race or sex. The current results replicate earlier reports of effects of poser race relative to background race on the LPC component of the ERP waveform. In addition, they indicate that an implicit race bias occurs regardless of participant's or poser's race and is not confined to a particular cultural context.

The relationship between self-reported animal fear and ERP modulation : Evidence for enhanced processing and fear of harmless invertebrates in snake- and spider-fearful individuals

The present study used ERPs to compare processing of fear-relevant (FR) animals (snakes and spiders) and non-fear-relevant (NFR) animals similar in appearance (worms and beetles). EEG was recorded from 18 undergraduate participants (10 females) as they completed two animal-viewing tasks that required simple categorization decisions. Participants were divided on a post hoc basis into low snake/spider fear and high snake/spider fear groups. Overall, FR animals were rated higher on fear and elicited a larger LPC. However, individual differences qualified these effects. Participants in the low fear group showed clear differentiation between FR and NFR animals on subjective ratings of fear and LPC modulation. In contrast, participants in the high fear group did not show such differentiation between FR and NFR animals. These findings suggest that the salience of feared-FR animals may generalize on both a behavioural and electro-cortical level to other animals of similar appearance but of a non-harmful nature.

Discovering the best feature extraction and selection algorithms for spontaneous facial expression recognition

Feature extraction and selection are critical processes in developing facial expression recognition (FER) systems. While many algorithms have been proposed for these processes, direct comparison between texture, geometry and their fusion, as well as between multiple selection algorithms has not been found for spontaneous FER. This paper addresses this issue by proposing a unified framework for a comparative study on the widely used texture (LBP, Gabor and SIFT) and geometric (FAP) features, using Adaboost, mRMR and SVM feature selection algorithms. Our experiments on the Feedtum and NVIE databases demonstrate the benefits of fusing geometric and texture features, where SIFT+FAP shows the best performance, while mRMR outperforms Adaboost and SVM. In terms of computational time, LBP and Gabor perform better than SIFT. The optimal combination of SIFT+FAP+mRMR also exhibits a state-of-the-art performance.