Hicham Atassi

A Speaker Independent Approach to the Classification of Emotional Vocal Expressions

The paper proposes a speaker independent procedure for classifying vocal expressions of emotion. The procedure is based on the splitting up of the emotion recognition process into two steps. In the first step, a combination of selected acoustic features is used to classify six emotions through a Bayesian Gaussian Mixture Model classifier (GMM). The two emotions that obtain the highest likelihood scores are selected for further processing in order to discriminate between them. For this purpose, a unique set of high-level acoustic features was identified using the Sequential Floating Forward Selection (SFFS) algorithm, and a GMM was used to separate between each couple of emotion. The mean classification rate is 81% with an improvement of 5% with respect to the most recent results obtained on the same database (75%).

Analysis of high-level features for vocal emotion recognition

The paper deals with the vocal emotion recognition which is a very important task for several applications in the field of human-machine interaction. There is a plenty of algorithms proposed up to date for this purpose that exploit different types of features and classifiers. Our previous work showed that high-level features perform very well in terms of emotion classification from speech. However, little attention has been paid so far to the statistical analysis of these features. For this reason the presented paper mainly focuses on the emotion recognition by using only high-level features. Two different emotional speech corpora were exploited in our experiments, namely the Berlin Database of Emotional Speech and the COST2102 Italian Database of Emotional Speech. Results showed that the best high-level features in terms of high discriminative power strongly differ among the databases considered on the first hand and among the emotions within each database on the second hand.

Emotional vocal expressions recognition using the COST 2102 italian database of emotional speech

The present paper proposes a new speaker-independent approach to the classification of emotional vocal expressions by using the COST 2102 Italian database of emotional speech. The audio records extracted from video clips of Italian movies possess a certain degree of spontaneity and are either noisy or slightly degraded by an interruption making the collected stimuli more realistic in comparison with available emotional databases containing utterances recorded under studio conditions. The audio stimuli represent 6 basic emotional states: happiness, sarcasm/irony, fear, anger, surprise, and sadness. For these more realistic conditions, and using a speaker independent approach, the proposed system is able to classify the emotions under examination with 60.7% accuracy by using a hierarchical structure consisting of a Perceptron and fifteen Gaussian Mixture Models (GMM) trained to distinguish within each pair (couple) of emotions under examination. The best features in terms of high discriminative power were selected by using the Sequential Floating Forward Selection (SFFS) algorithm among a large number of spectral, prosodic and voice quality features. The results were compared with the subjective evaluation of the stimuli provided by human subjects.

Generation of biometric based unique digital watermark from iris image

This paper proposes a proficient digital watermark generation technique from biometric data which will be unique and can be logically owned to prove ownership. The biometric pattern of iris is used to generate the digital watermark that has a clear stamp of ownership. The generated watermark has been studied for uniqueness and identification and has been used to watermark audio signals. Dither modulation quantization is applied on the singular values of Singular Value Decomposition domain for embedding the watermark. Experimental results indicates that the watermark can survive the signal processing attacks such as Gaussian noise corruption, re-sampling, re-quantization, cropping, and compression and maintain the perceptual properties of the host signal and hence satisfies the design requirements of digital watermarking. The extracted biometric based watermark was uniquely identified under signal processing attacks.

A novel clinical expert system for chest pain risk assessment

Rapid access chest pain clinics (RACPC) enable clinical risk assessment, investigation and arrangement of a treatment plan for chest pain patients without a long waiting list. RACPC Clinicians often experience difficulties in the diagnosis of chest pain due to the inherent complexity of the clinical process and lack of comprehensive automated diagnostic tools. To date, various risk assessment models have been proposed, inspired by the National Institute of Clinical Excellence (NICE) guidelines to provide clinical decision support mechanism in chest pain diagnosis. The aim of this study is to help improve the performance of RACPC, specifically from the clinical decision support perspective. The study cohort comprises of 632 patients suspected of cardiac chest pain. A retrospective data analysis of the clinical studies evaluating 14 risk factors for chest pain patients was performed for the development of RACPC specific risk assessment models to distinguish between cardiac and non cardiac chest pain. In the first phase, a novel binary classification model was developed using a Decision Tree algorithm in conjunction with forward and backward selection wrapping techniques. Secondly, a logistic regression model was trained using all of the given variables combined with forward and backward feature selection techniques to identify the most significant features. The new models have resulted in very good predictive power, demonstrating general performance improvement compared to a state-of-the-art prediction model.

A novel cardiovascular decision support framework for effective clinical risk assessment

The aim of this study is to help improve the diagnostic and performance capabilities of Rapid Access Chest Pain Clinics (RACPC), by reducing delay and inaccuracies in the cardiovascular risk assessment of patients with chest pain by helping clinicians effectively distinguish acute angina patients from those with other causes of chest pain. Key to our new approach is (1) an intelligent prospective clinical decision support framework for primary and secondary care clinicians, (2) learning from missing/impartial clinical data using Bernoulli mixture models and Expectation Maximisation (EM) techniques, (3) utilisation of state-of-the-art feature section, pattern recognition and data mining techniques for the development of intelligent risk prediction models for cardiovascular patients. The study cohort comprises of 632 patients suspected of cardiac chest pain. A retrospective data analysis of the clinical studies evaluating clinical risk factors for chest pain patients was performed for the development of RACPC specific risk assessment models to distinguish between cardiac and non cardiac chest pain. A comparative analysis case study of machine learning methods was carried out for predicting RACPC clinical outcomes using real patient data acquired from Raigmore Hospital in Inverness, UK. The proposed framework was also validated using the University of Clevelands Heart Disease dataset which contains 76 attributes, but all published experiments refer to using a subset of 14 of them. Experiments with the Cleveland database (based on 18 clinical features of 270 patients) were concentrated on attempting to distinguish the presence of heart disease (values 1, 2, 3, 4) from absence (value 0). The new clinical models, having been evaluated in clinical practice, resulted in very good predictive power, demonstrating general performance improvement over benchmark multivariate statistical classifiers. As part of these case studies, various online RACPC risk assessment prototypes have been developed which are being deployed in the clinical setting (NHS Highland) for clinical trial purposes.

Detection of chorus from an audio clip using dynamic time warping algorithm

This paper describes a method to detect repeating segments in an audio signal by using dynamic time warping algorithm. The proposed framework extracts features from frames of the audio by Mel frequency cepstral coefficients. The features extracted from the audio clip of the chorus were matched against the features of the whole clip by dynamic time warping. The number of matches found was determined by self similarity matrix. The experimental results indicate that the minimum distance matches between query and reference clip is successfully achieved. The proposed scheme was tested in a database of audio signals and the experimental results are encouraging. The proposed scheme was implemented and tested using a database of audio signals with accuracy up to 98%.