page d'accueil   sommaire  

Abdellah KACHA


coordonnées


Ecole polytechnique de Bruxelles
Abdellah KACHA
tel 02 650 30 85, akacha@ulb.ac.be
Campus du Solbosch
CP165/51, avenue F.D. Roosevelt 50, 1050 Bruxelles



unités de recherche


Traitement des Signaux (LIST/TS) [Signal Processing Group] (LIST/TS)



projets


ECLIPSE - Functional assessment of disordered voices and real-time monitoring under field conditions - Evaluation fonctionnelle clinique des pathologies vocales et suive embarqué [ECLIPSE - Functional assessment of disordered voices and real-time monitoring under field conditions]
The objective is the development of novel speech analysis methods that enable describing normal and pathological voices reliably and validly on the base of connected speech (as opposed to sustained vowels, which are the norm). High-speed imaging of the vibrating vocal folds is combined with acoustic analysis of sustained sounds to provide for clinical use multiple modalities that complement each other. The overall objective is the development of a clinical workstation running multiple analysis algorithms and the building of a portable device that enables monitoring speakers under field conditions. This project is a common project with the University of Mons and the Catholic University of Louvain. Cette recherche consiste à développer des nouvelles méthodes d'analyse du signal de parole permettant de caractériser de manière objective les voix pathologiques. Les techniques sont applicables à la parole continue contrairement aux méthodes actuelles qui sont limitées à l'analyse des voyelles soutenues. En parallèle, un traitement d'images des cordes vocales, prises par caméra haute vitesse, est mis à profit pour mettre en place une analyse multimodale et d'en d'améliorer l'efficacité. L'objectif global du projet est de développer une station de travail clinique incluant les logiciels d'analyse dédiés ainsi que la réalisation d'un dispositif portable permettant un suivi en continu des locuteurs à risque. Ce projet est réalisé en collaboration avec l'Université de Mons et l'Université Catholique de Louvain. [The objective is the development of novel speech analysis methods that enable describing normal and pathological voices reliably and validly on the base of connected speech (as opposed to sustained vowels, which are the norm). High-speed imaging of the vibrating vocal folds is combined with acoustic analysis of sustained sounds to provide for clinical use multiple modalities that complement each other. The overall objective is the development of a clinical workstation running multiple analysis algorithms and the building of a portable device that enables monitoring speakers under field conditions. This project is a common project with the University of Mons and the Catholic University of Louvain.]

Acoustic-to-articulatory mapping - Inversion acoustico-articulatoire [Acoustic-to-articulatory mapping]
Acoustic-to-articulatory mapping is the inference from the speech signal of the vocal tract shape from the glottis to the lips. A first method has consisted in recovering the vocal tract shape by means of the formant frequencies and constraints at the lips. A second method involves the inference of the vocal tract shape based on a simulation of the speech spectrum, anti-formants, sources and losses included. The objectives are the extraction of the formant frequencies by means of the inferred tract shapes as well as the representation of the speech signal in terms of morphological rather than acoustic features. L'inversion acoustico-articulatoire consiste à retrouver la forme du conduit vocal, depuis les cordes vocales jusqu'aux lèvres, à partir du signal acoustique. Une première étude a concerné la reconstruction de la fonction d'aire du conduit à partie de la connaissance des fréquences des premiers formants, en ajoutant éventuellement une contrainte sur l'ouverture aux lèvres. Une deuxième approche est basée sur une modélisation globale du spectre incluant également l'effet des anti-formants ainsi que de la source et des pertes. Les objectifs sont l'extraction des fréquences des formants ainsi que la représentation du signal de parole en termes d'indices morphologiques plutôt qu'acoustiques. [Acoustic-to-articulatory mapping is the inference from the speech signal of the vocal tract shape from the glottis to the lips. A first method has consisted in recovering the vocal tract shape by means of the formant frequencies and constraints at the lips. A second method involves the inference of the vocal tract shape based on a simulation of the speech spectrum, anti-formants, sources and losses included. The objectives are the extraction of the formant frequencies by means of the inferred tract shapes as well as the representation of the speech signal in terms of morphological rather than acoustic features.]

Speech Signal Analysis for Voice Disorders Characterization - Analyse du signal de parole pour la caractérisation des troubles de la voix [Speech Signal Analysis for Voice Disorders Characterization]
The analysis of speech is often involved in the evaluation of voice and the discrimination between disphonic and normophonic speakers. Most of the time, acoustic features of voice disorders are obtained from sustained vowels. However, sustaining speech sounds is considered to be less informative than speaking continuously. The objective of the project therefore is to monitor the voice timbre and vocal perturbations in connected speech. Vocal disperiodicities obtained are summarised by means of vocal features that report speech properties that are clinically relevant with regard to the assessment of a speaker's voice. The description of voice disorders owing to peripheral causes are also considered. L'analyse acoustique du signal de parole est souvent utilisée dans l'évaluation et la discrimination des voix pathologiques. Les mesures acoustiques utilisées pour caractériser les troubles de la voix sont extraites de voyelles soutenues. Cependant, la représentation basée sur les voyelles soutenues n'est pas toujours valide. L'estimation de paramètres à partir de la parole connectée semble plus adéquate. L'objectif de ce projet est le développement de méthodes permettant le suivi d'un locuteur au cours de la lecture d'un texte de longue durée. L'utilisation de méthodes applicables à des signaux non stationnaires est aussi envisagée. Le bruit estimé par les différentes méthodes permet de déterminer les indices pertinents utilisés dans la caractérisation des troubles de la voix du locuteur. La caractérisation de certains troubles vocaux dus à un problème neurologique périphérique, comme les voix dysarthriques, est aussi examinée. [The analysis of speech is often involved in the evaluation of voice and the discrimination between disphonic and normophonic speakers. Most of the time, acoustic features of voice disorders are obtained from sustained vowels. However, sustaining speech sounds is considered to be less informative than speaking continuously. The objective of the project therefore is to monitor the voice timbre and vocal perturbations in connected speech. Vocal disperiodicities obtained are summarised by means of vocal features that report speech properties that are clinically relevant with regard to the assessment of a speaker's voice. The description of voice disorders owing to peripheral causes are also considered. ]

Détection et prédiction de crises d'épilepsie [On-line detection and prediction of epileptic seizures]
L'objectif de cette recherche est de modéliser à l'aide de circuits oscillants l'activité électrique neuronale et son comportement pathologique lors des crises d'épilepsie. Cela permettra de mieux comprendre le processus qui amène à la crise et de connaître les paramètres qui sont les marqueurs du changement dans la dynamique de l'EEG. Des algorithmes de détection et de prédiction seront alors développés sue base des informations issues de la modélisation. [Electroencephalography is a useful tool for physicians. It can be used for diagnostic purposes in many brain's pathologies as epilepsy, sleep troubles, etc. Epilepsy is the most current pathology in this field and concerns about 1% of the population. One of the characteristics of epilepsy is the presence of repetitive seizures. Clinically, these seizures can take different forms: from muscular moves to critical convulsions called 'Grand Mal' or tonic-clonic seizure. In EEG signals, an epileptic seizure is characterized by high-voltage and rhythmic EEG waveforms. High-voltage signals show a phenomenon of high synchronous activities of the brain or a part of it. Epilepsy can be focal (or partial) if the seizure is reduced to a part of the brain, called the focus. If the entire brain is affected, the epilepsy is said to be generalized. The automatic detection of epileptic seizure is an important purpose for specialist's diagnostic, because it could save the specialist time by pointing out the seizure periods. Furthermore, another interesting challenge in the field of EEG signal processing is the ability of predicting a seizure. It could enable in the future to avoid seizures of the patients by means of different techniques. Several linear and nonlinear features have been investigated in the literature. But at the present time, it's difficult to say if an epileptic seizure is reliably predictable. In the literature, different kinds of models have been proposed to simulate EEG signals. They use neurophysiologic mechanisms and are called 'mean-field models' because of their macroscopic approach. These models are able to represent epileptic-like EEG signals as well as normal EEG signals. They strongly suggest that epileptic activity is related with the instability or quasi-instability of the system leading to rhythmic activities. Investigating this feature of epileptic process in EEG signals could permit to better understand this process and to provide a feature for characterizing the epileptic seizure in computer-based monitoring of EEG. ]

ASPI - Audiovisual to Articulatory Speech Inversion. European FET Project. [ASPI - Audiovisual to Articulatory Speech Inversion. European FET Project.]
ASPI has been an European FET project involving LORIA (France), KTH (Sweden), TUA (Greece), ENST (France) and ULB (Belgium). Audiovisual-to-articulatory inversion consists in recovering the vocal tract shape (from the vocal folds to the lips) dynamics from the acoustical speech signal, supplemented by image analysis of the speaker's face. The design of audiovisual-to-articulatory inversion involves two kinds of interdependent tasks. The first is the development of inversion methods that successfully answer the main acknowledged difficulties (non-unicity of inverse solution, a possible lack of phonetic relevancy of inverse solutions, impossibility of using standard spectral data), and the second is the construction of an articulatory database that comprises dynamic images of the vocal tract together with the speech signal uttered, and that for several male and female speakers. [ASPI has been an European FET project involving LORIA (France), KTH (Sweden), TUA (Greece), ENST (France) and ULB (Belgium). Audiovisual-to-articulatory inversion consists in recovering the vocal tract shape (from the vocal folds to the lips) dynamics from the acoustical speech signal, supplemented by image analysis of the speaker's face. The design of audiovisual-to-articulatory inversion involves two kinds of interdependent tasks. The first is the development of inversion methods that successfully answer the main acknowledged difficulties (non-unicity of inverse solution, a possible lack of phonetic relevancy of inverse solutions, impossibility of using standard spectral data), and the second is the construction of an articulatory database that comprises dynamic images of the vocal tract together with the speech signal uttered, and that for several male and female speakers. ]



disciplines et mots clés déclarés


Electromagnétisme - analyse du signal Informatique médicale Ingénierie biomédicale Phonétique Sciences de l'ingénieur

analyse de la parole analyse multimodale bruit vocal crises d'épilepsie détection et prédiction dispositif portable inversion acoustique inversion audiovisuelle signal de parole signaux non stationnaires traitement du signal traitement du signal de parole traitement du signal eeg troubles de la voix voix pathologiques