Reda A. El-Khoribi

Audio-Visual Speech Recognition for People with Speech Disorders

recognition of disorder people is a difficult task due to the lack of motor-control of the speech articulators. Multimodal speech recognition can be used to enhance the robustness of disordered speech. This paper introduces an automatic speech recognition system for people with dysarthria speech disorder based on both speech and visual components. The Mel-Frequency Cepestral Coefficients (MFCC) is used as features representing the acoustic speech signal. For the visual counterpart, the Discrete Cosine Transform (DCT) Coefficients are extracted from the speakers mouth region. Face and mouth regions are detected using the Viola-Jones algorithm. The acoustic and visual input features are then concatenated on one feature vector. Then, the Hidden Markov Model (HMM) classifier is applied on the combined feature vector of acoustic and visual components. The system is tested on isolated English words spoken by disorder speakers from UA-Speech data. Results of the proposed system indicate that visual features are highly effective and can improve the accuracy to reach 7.91% for speaker dependent experiments and 3% for speaker independent experiments.

Classification of motor imagery tasks with LS-SVM in EEG-based self-paced BCI

Motivated by the need to deal with critical disorders that involve death of neurons, such as Amyotrophic Lateral Sclerosis (ALS) and brainstem stroke, interpretation of the brains Motor Imagery (MI) activities is highly needed. Brain signals can be translated into control commands. Electroencephalography (EEG) is considered in this work, EEG is a low-cost non-invasive technique. A big challenge is faced due to the poor signal-to-noise ratio of EEG signals. The dataset used in this work is based on asynchronous or self-paced motor imagery problem. The used self-paced Brain Computer Interface (BCI) problem poses a considerable challenge by introducing an additional class, a relax class, or non-intentional control periods that are not included in the training set and should be classified. In this work, a number of subject dependent parameters and their values are determined. These parameters are: the best frequency range, the best Common Spatial Pattern (CSP) channels, and the number of these CSP channels. System parameters are determined dynamically in the offline training phase. Energy based features are extracted afterwards from the best selected signals. The Least-Squares Support Vector Machine (LS-SVM) classifier is used as a classification back end. Results of the proposed system show superiority over the previously introduced systems in terms of the Mean Square Error (MSE) when tested on the Berlin BCI (BBCI) competition IV dataset 1.

An integrated classification method for brain computer interface system

A channel of communication for both human brain and computer system is provided via a system called Brain Computer Interface (BCI). The vital aim of BCI research is to develop a system that helps the disabled people to interact with other persons and allows their interaction with the external environments or as an additional man-machine interaction channel for healthy users. Different techniques have been developed in the literature for the classification of brain signals. The purpose of this work is to deveolp a novel method of analyzing the EEG signals. We have used high pass filter to remove artifacts, DWT algorithms for feature extraction and features like Mean Absolute Value, Root Mean Square, and Simple Square Integral are used. The neural network algorithm is used to find the correct class label for EEG signal after clustering the feature vectors using K-Nearest Neighbor algorithm. It has been depicted from results that the proposed integrated technique outperforms a better performance than methods mentioned in literature.

Trajectory Learning Using Principal Component Analysis

Robots are increasingly used in numerous life applications. Therefore, humans are looking forward to create productive robots. Robot learning is the process of obtaining additional information to accomplish an objective configuration. Moreover, robot learning from demonstration is to guide the robot the way to perform a particular task derived from human directions. Traditionally, modeling the demonstrated data was applied on discrete data which would result in learning outcome distortions. So as to overcome such distortion, preprocessing of the raw data is necessary. In this paper, trajectory learning from demonstration scheme is proposed. In our proposed scheme, the raw data are initially preprocessed by employing the principal component analysis algorithm. We experimentally compare our proposed scheme with the most recent proposed schemes. It is found that the proposed scheme is capable of increasing the efficiency by minimizing the error in comparison to the other recent work with significant reduced computational cost.

An improved technique for LIDAR data reduction

Light detection and ranging (LIDAR) is a technology of remote imaging technologies. Currently, it is the most important technology for accruing elevation points with a high density in the form of digital elevation model (DEM) construction. However, the high-density data leads to time and memory consumption problems during data processing. In this paper, we depend on radial basis function (RBF) with Gaussian interpolation method to carry out LIDAR data reduction by select the most important points from the unprocessed data to remain the constructed DEMs with high accuracy as possible. Comparing the results with respect to the accuracy using Structural Similarity Index (SSIM) with Multiquadric and TPS interpolation methods. The results showing that Gaussian method is the most accurate method with 5.49% regardless each Multiquadric and TPS methods.