Exploring visual features through Gabor representations for facial expression detection

Gabor representations have been widely used in facial analysis (face recognition, face detection and facial expression detection) due to their biological relevance and computational properties. Two popular Gabor representations used in literature are: 1) Log-Gabor and 2) Gabor energy filters. Even though these representations are somewhat similar, they also have distinct differences as the Log-Gabor filters mimic the simple cells in the visual cortex while the Gabor energy filters emulate the complex cells, which causes subtle differences in the responses. In this paper, we analyze the difference between these two Gabor representations and quantify these differences on the task of facial action unit (AU) detection. In our experiments conducted on the Cohn-Kanade dataset, we report an average area underneath the ROC curve (A`) of 92.60% across 17 AUs for the Gabor energy filters, while the Log-Gabor representation achieved an average A` of 96.11%. This result suggests that small spatial differences that the Log-Gabor filters pick up on are more useful for AU detection than the differences in contours and edges that the Gabor energy filters extract.

Cascading appearance-based features for visual voice activity detection

The detection of voice activity is a challenging problem, especially when the level of acoustic noise is high. Most current approaches only utilise the audio signal, making them susceptible to acoustic noise. An obvious approach to overcome this is to use the visual modality. The current state-of-the-art visual feature extraction technique is one that uses a cascade of visual features (i.e. 2D-DCT, feature mean normalisation, interstep LDA). In this paper, we investigate the effectiveness of this technique for the task of visual voice activity detection (VAD), and analyse each stage of the cascade and quantify the relative improvement in performance gained by each successive stage. The experiments were conducted on the CUAVE database and our results highlight that the dynamics of the visual modality can be used to good effect to improve visual voice activity detection performance.

Optimising figure of merit for phonetic spoken term detection

This paper introduces a novel technique to directly optimise the Figure of Merit (FOM) for phonetic spoken term detection. The FOM is a popular measure of sTD accuracy, making it an ideal candiate for use as an objective function. A simple linear model is introduced to transform the phone log-posterior probabilities output by a phe classifier to produce enhanced log-posterior features that are more suitable for the STD task. Direct optimisation of the FOM is then performed by training the parameters of this model using a non-linear gradient descent algorithm. Substantial FOM improvements of 11% relative are achieved on held-out evaluation data, demonstrating the generalisability of the approach.

Fast and accurate phonetic spoken term detection

For the first time in human history, large volumes of spoken audio are being broadcast, made available on the internet, archived, and monitored for surveillance every day. New technologies are urgently required to unlock these vast and powerful stores of information. Spoken Term Detection (STD) systems provide access to speech collections by detecting individual occurrences of specified search terms. The aim of this work is to develop improved STD solutions based on phonetic indexing. In particular, this work aims to develop phonetic STD systems for applications that require open-vocabulary search, fast indexing and search speeds, and accurate term detection. Within this scope, novel contributions are made within two research themes, that is, accommodating phone recognition errors and, secondly, modelling uncertainty with probabilistic scores. A state-of-the-art Dynamic Match Lattice Spotting (DMLS) system is used to address the problem of accommodating phone recognition errors with approximate phone sequence matching. Extensive experimentation on the use of DMLS is carried out and a number of novel enhancements are developed that provide for faster indexing, faster search, and improved accuracy. Firstly, a novel comparison of methods for deriving a phone error cost model is presented to improve STD accuracy, resulting in up to a 33% improvement in the Figure of Merit. A method is also presented for drastically increasing the speed of DMLS search by at least an order of magnitude with no loss in search accuracy. An investigation is then presented of the effects of increasing indexing speed for DMLS, by using simpler modelling during phone decoding, with results highlighting the trade-off between indexing speed, search speed and search accuracy. The Figure of Merit is further improved by up to 25% using a novel proposal to utilise word-level language modelling during DMLS indexing. Analysis shows that this use of language modelling can, however, be unhelpful or even disadvantageous for terms with a very low language model probability. The DMLS approach to STD involves generating an index of phone sequences using phone recognition. An alternative approach to phonetic STD is also investigated that instead indexes probabilistic acoustic scores in the form of a posterior-feature matrix. A state-of-the-art system is described and its use for STD is explored through several experiments on spontaneous conversational telephone speech. A novel technique and framework is proposed for discriminatively training such a system to directly maximise the Figure of Merit. This results in a 13% improvement in the Figure of Merit on held-out data. The framework is also found to be particularly useful for index compression in conjunction with the proposed optimisation technique, providing for a substantial index compression factor in addition to an overall gain in the Figure of Merit. These contributions significantly advance the state-of-the-art in phonetic STD, by improving the utility of such systems in a wide range of applications.

Discriminative optimisation of the figure of merit for phonetic spoken term detection

This work proposes to improve spoken term detection (STD) accuracy by optimising the Figure of Merit (FOM). In this article, the index takes the form of phonetic posterior-feature matrix. Accuracy is improved by formulating STD as a discriminative training problem and directly optimising the FOM, through its use as an objective function to train a transformation of the index. The outcome of indexing is then a matrix of enhanced posterior-features that are directly tailored for the STD task. The technique is shown to improve the FOM by up to 13% on held-out data. Additional analysis explores the effect of the technique on phone recognition accuracy, examines the actual values of the learned transform, and demonstrates that using an extended training data set results in further improvement in the FOM.

Visual front-end wars : Viola-Jones face detector vs Fourier Lucas-Kanade

The performance of visual speech recognition (VSR) systems are significantly influenced by the accuracy of the visual front-end. The current state-of-the-art VSR systems use off-the-shelf face detectors such as Viola- Jones (VJ) which has limited reliability for changes in illumination and head poses. For a VSR system to perform well under these conditions, an accurate visual front end is required. This is an important problem to be solved in many practical implementations of audio visual speech recognition systems, for example in automotive environments for an efficient human-vehicle computer interface. In this paper, we re-examine the current state-of-the-art VSR by comparing off-the-shelf face detectors with the recently developed Fourier Lucas-Kanade (FLK) image alignment technique. A variety of image alignment and visual speech recognition experiments are performed on a clean dataset as well as with a challenging automotive audio-visual speech dataset. Our results indicate that the FLK image alignment technique can significantly outperform off-the shelf face detectors, but requires frequent fine-tuning.

How everyday language can and will boost effective information retrieval

Typing 2 or 3 keywords into a browser has become an easy and efficient way to find information. Yet, typing even short queries becomes tedious on ever shrinking (virtual) keyboards. Meanwhile, speech processing is maturing rapidly, facilitating everyday language input. Also, wearable technology can inform users proactively by listening in on their conversations or processing their social media interactions. Given these developments, everyday language may soon become the new input of choice. We present an information retrieval (IR) algorithm specifically designed to accept everyday language. It integrates two paradigms of information retrieval, previously studied in isolation; one directed mainly at the surface structure of language, the other primarily at the underlying meaning. The integration was achieved by a Markov machine that encodes meaning by its transition graph, and surface structure by the language it generates. A rigorous evaluation of the approach showed, first, that it can compete with the quality of existing language models, second, that it is more effective the more verbose the input, and third, as a consequence, that it is promising for an imminent transition from keyword input, where the onus is on the user to formulate concise queries, to a modality where users can express more freely, more informal, and more natural their need for information in everyday language.

Bayesian and Decision Tree Approaches for Pattern Recognition Including Feature Measurement Costs

The minimum cost classifier when general cost functionsare associated with the tasks of feature measurement and classification is formulated as a decision graph which does not reject class labels at intermediate stages. Noting its complexities, a heuristic procedure to simplify this scheme to a binary decision tree is presented. The optimizationof the binary tree in this context is carried out using ynamicprogramming. This technique is applied to the voiced-unvoiced-silence classification in speech processing.

Analysis of ECG by multipulse excited linear prediction model

He propose a new time domain method for efficient representation of the KCG and delineation of its component waves. The method is based on the multipulse Linear prediction (LP) coding which is being widely used in speech processing. The excitation to the LP synthesis filter consists of a few pulses defined by their locations and amplitudes. Based on the amplitudes and their distribution, the pulses are suitably combined to delineate the component waves. Beat to beat correlation in the ECG signal is used in QRS periodicity prediction. The method entails a data compression of 1 in 6. The method reconstructs the signal with an NMSE of less than 5%.

Fast acoustic likelihood computation using low-rank matrix approximation

Acoustic modeling using mixtures of multivariate Gaussians is the prevalent approach for many speech processing problems. Computing likelihoods against a large set of Gaussians is required as a part of many speech processing systems and it is the computationally dominant phase for LVCSR systems. We express the likelihood computation as a multiplication of matrices representing augmented feature vectors and Gaussian parameters. The computational gain of this approach over traditional methods is by exploiting the structure of these matrices and efficient implementation of their multiplication.In particular, we explore direct low-rank approximation of the Gaussian parameter matrix and indirect derivation of low-rank factors of the Gaussian parameter matrix by optimum approximation of the likelihood matrix. We show that both the methods lead to similar speedups but the latter leads to far lesser impact on the recognition accuracy. Experiments on a 1138 word vocabulary RM1 task using Sphinx 3.7 system show that, for a typical case the matrix multiplication approach leads to overall speedup of 46%. Both the low-rank approximation methods increase the speedup to around 60%, with the former method increasing the word error rate (WER) from 3.2% to 6.6%, while the latter increases the WER from 3.2% to 3.5%.