Electroglottography (EGG) is a noninvasive technique which allows accurate measurements of vocal folds dynamics and perturbations during speech. It has been widely used in medical diagnosis and monitoring of several laryngeal pathologies, but its use in neurological disorders has been very limited. This paper presents the first study on EGG in early stages of Parkinson's disease (PD) and an atypical parkinsonian disorder, multiple system atrophy (MSA). Our first objective was to investigate whether EGG can reveal distinctive dysphonia features that could help in the challenging early differential diagnosis between PD and MSA-P, the parkinsonian variant of MSA. The second objective was to verify some hypothesis on early markers of PD drawn from speech-alone acoustic analysis. For MSA-P patients, our analysis revealed a reduced open quotient and confirmed excessive pitch variation. The analysis also mitigated some hypothesis on dysphonia in early stages of PD. © 2024 International Speech Communication Association. All rights reserved.

The study of coastal upwelling through the analysis of sea surface temperature (SST) satellite imagery has been a valuable approach because of its efficiency and practicality. Building on inception and residual structures, we introduce IncepResup-Net, a novel deep learning model for identifying upwelling regions along Morocco’s Atlantic coast. This model effectively addresses limitations in recent methods targeting the same upwelling system and outperforms them by more accurately detecting true upwelling areas, thereby minimizing false positives. Applied to SST data spanning from 2000 to 2022, IncepResup-Net demonstrates superior performance over traditional and contemporary deep learning models, marked by its precise segmentation capabilities and robustness in real-world detection scenarios. Our findings highlight the model’s effectiveness in leveraging SST imagery for upwelling detection, establishing a new benchmark in the application of deep learning within geoscience and remote sensing fields.

Acoustic realisation of the working vowel space has been widely studied in Parkinson's disease (PD).However, it has never been studied in atypical parkinsonian disorders (APD).The latter are neurodegenerative diseases which share similar clinical features with PD, rendering the differential diagnosis very challenging in early disease stages.This paper presents the first contribution in vowel space analysis in APD, by comparing corner vowel realisation in PD and the parkinsonian variant of Multiple System Atrophy (MSA-P).Our study has the particularity of focusing exclusively on early stage PD and MSA-P patients, as our main purpose was early differential diagnosis between these two diseases.We analysed the corner vowels, extracted from a spoken sentence, using traditional vowel space metrics.We found no statistical difference between the PD group and healthy controls (HC) while MSA-P exhibited significant differences with the PD and HC groups.We also found that some metrics conveyed complementary discriminative information.Consequently, we argue that restriction in the acoustic realisation of corner vowels cannot be a viable early marker of PD, as hypothesised by some studies, but it might be a candidate as an early hypokinetic marker of MSA-P (when the clinical target is discrimination between PD and MSA-P).

Parkinson's disease (PD) and the parkinsonian variant of Multiple System Atrophy (MSA-P) are two neurodegenerative diseases which share similar clinical features, particularly in early disease stages. The differential diagnosis can be thus very challenging. Dysarthria is known to be a frequent and early clinical feature of PD and MSA. It can be thus used as a vehicle to provide a vocal biomarker which could help in the differential diagnosis. In particular, distortion of consonants is known to be a frequent impairment in these diseases. The aim of this study is to investigate distinctive patterns in the distortion of voiced obstruents (plosives and fricatives). It is the first study which attempts to examine such distortions in the French language for the purpose of the differential diagnosis between PD and MSAP (and among the very few studies if we consider all languages). We carry out a perceptual and objective analysis of voiced obstruents extracted from isolated pseudo-words initials. We first show that devoicing is a significant impairment which predominates in MSA-P. We then show that voice onset time (VOT) of voiced plosives (prevoicing duration) can be a complementary feature to improve the accuracy in discrimination between PD and MSA-P.

Analysis and study of coastal upwelling using sea surface temperature (SST) satellite images is a common procedure because of its coast effectiveness (economic, time, frequency, and manpower). Developing on the Ekman theory, we propose a robust method to identify the upwelling regions along the north-west African margin. The proposed method comes to overcome the issues encountered in a recent method devoted for the same purpose and for the same upwelling system. Afterward, we show how our method can serve as a framework to study and monitor the spatio-temporal variability of the upwelling phenomenon in the studied region.

The region along the North-West African coast (20°N to 36°N and 4°W to 19°W) is characterized by a persistent and variable upwelling phenomenon almost all year round. In this article, the upwelling features are investigated using an algorithm dedicated to delimit the upwelling area from thermal and biological satellite observations. This method has been developed specifically for sea-surface temperature (SST) images, since they present a high latitudinal variation, which is not present in chlorophyll-a concentration images. Developing on the proposed approach, the spatial and temporal variations of the main physical and biological upwelling patterns are studied. Moreover, a study on the upwelling dynamics, which explores the interplay between the upwelling spatiotemporal extents and intensity, is presented, based on a 14-year time archive of weekly SST and chlorophyll-a concentration data.

Parkinsonism refers to Parkinson's Disease (PD) and Atypical Parkinsonian Syndromes (APS), such as Progressive Supranuclear Palsy (PSP) and Multiple System Atrophy (MSA). Discrimination between PD and APS and within APS groups in early disease stages is a very challenging task. Interestingly, speech disorder is frequently an early and prominent clinical feature of both PD and APS. This renders speech/voice analysis a promising tool for the development of an objective marker to assist neurologists in their diagnosis. This paper is a continuation of a recent work on speech-based differential diagnosis within APS. We address the difficult problem of defining disease-specific speech features which is crucial in the perspective of early differential diagnosis. We investigate this problem by considering the constraint that only a small amount of training data can be available in this setting. To do so, we perform univariate statistical analysis followed by a supervised learning that forces the designed new features to be 1-dimensional. We carry out experiments using speech recordings of MSA and PSP patients. We show that linear classification models allow the definition of new scalar variables which can be considered as speech features which are specific to each disease, MSA and PSP.

Parkinsonism refers to Parkinsons disease (PD) and Atypical parkinsonian syndromes (APS). Speech disorder is a common and early symptom in Parkinsonism which makes speech analysis a very important research area for the purpose of early diagnosis. Most of research have however focused on discrimination between PD and healthy controls. Such research does not take into account the fact that PD and APS syndromes are very similar in early disease stages. The main problem that has to be addressed first is then differential diagnosis: discrimination between PD and APS and within APS. This paper is a continuation of an earlier pioneer work in differential diagnosis where we mostly address the machine learning problem due to the small amount of training data. We show that classical linear and generalized linear models can provide interpretable and robust classifiers in term of accuracy and generalization ability.

The aim of this work is to automatically identify and extract the upwelling area in the coastal ocean of Morocco using the satellite observation of chlorophyll concentration. The algorithm starts by the application of FCM algorithm for the purpose of finding regions of homogeneous concentration of the chlorophyll, resulting in c-partitioned labeled images. A region-growing algorithm is then used to filter out the noisy structures in the offshore waters not belonging to the upwelling regions. The proposed methodology has been validated by an oceanographer and tested over a database of 166 weekly Sea Surface chlorophyll data. The region of interst cover the southern part of Moroccan atlantic coast spanning from the years 2007 to 2012.

Classical pitch-based perturbation measures, such as Jitter and Shimmer, are generally based on detection algorithms of pitch marks which assume the existence of a periodic pitch pattern and/or rely on the linear source-filter speech model. While these assumptions can hold for normal speech, they are generally not valid for pathological speech. The latter can indeed present strong aperiodicity, nonlinearity and turbulence/noise. Recently, we introduced on a novel nonlinear algorithm for Glottal Closure Instants (GCI) detection which has the strong advantage of not making such assumptions. In this paper, we use this new algorithm to compute standard pitch-based perturbation measures and compare its performances to the widely used tool PRAAT. We address the task of classification between normal and pathological speech, and carry out the experiments using the popular MEEI database. The results show that our algorithm leads to significantly higher classification accuracy than PRAAT. Moreover, some important statistical features become significantly discriminative, while they are meaningless when using PRAAT (in the sense that they have almost no discrimination power).

An unsupervised classification method is developed for the coarse segmentation of Moroccan coastal upwelling using the sea surface temperature (SST) satellite images. The algorithm is started with the generation of a c-partitioned labeled image using Otsu's method for the purpose of finding regions of homogenous temperatures. Then, two well-known validity indices are used to select the c-partition that best reproduces the shape of upwelling area. A region-growing algorithm is developed that is used to remove the noisy structures in the offshore waters not belonging to the upwelling area. The algorithm is used to provide a seasonal variability of upwelling activity in the southern Moroccan Atlantic coast using 70 SST images of the years 2007 and 2008. The performance of the proposed methodology has been validated by an oceanographer, showing its effectiveness for automatic delimitation of the Moroccan upwelling region.

International audience

This paper presents a novel algorithm for automatic detection of Glottal Closure Instants (GCI) from the speech signal. Our approach is based on a novel multiscale method that relies on precise estimation of a multiscale parameter at each time instant in the signal domain. This parameter quantifies the degree of signal singularity at each sample from a multi-scale point of view and thus its value can be used to classify signal samples accordingly. We use this property to develop a simple algorithm for detection of GCIs and we show that for the case of clean speech, our algorithm performs almost as well as a recent state-of-the-art method. Next, by performing a comprehensive comparison in presence of 14 different types of noises, we show that our method is more accurate (particularly for very low SNRs). Our method has lower computational times compared to others and does not rely on an estimate of pitch period or any critical choice of parameters.

INTERSPEECH 2012

In an earlier work, we proposed a novel phonetic segmentation method based on speech analysis under the Microcanonical Multiscale Formalism (MMF). The latter relies on the computation of local geometrical parameters, singularity exponents (SE). We showed that SE convey valuable information about the local dynamics of speech that can readily and simply used to detect phoneme boundaries. By performing error analysis of our original algorithm, in this paper we propose a 2-steps technique which better exploits SE to improve the segmentation accuracy. In the first step, we detect the boundaries of the original signal and of a low-pass filtred version, and we consider the union of all detected boundaries as candidates. In the second step, we use a hypothesis test over the local SE distribution of the original signal to select the final boundaries. We carry out a detailed evaluation and comparison over the full training set of the TIMIT database which could be useful to other researchers for comparison purposes. The results show that the new algorithm not only outperforms the original one, but also is significantly much more accurate than state-of-the-art ones.

International audience

International audience

We present a comparative study of several SVM speaker verification (SV) systems based on sequence kernels: the GMM-supervectors kernel, the fisher kernel, the Generalized linear discriminant sequence (GLDS) kernel, our feature space normalized sequence (FSNS) kernel and a ldquonovelrdquo sequence kernel in SV, the correlation kernel. We also compare these SVM systems to the conventional generative UBM-GMM. We carry out experiments on the NIST'2005 SRE evaluation set. The results show that the FSNS system yields comparable performances to UBM-GMM and significantly outperforms GLDS. They also show that the GMM-supervectors system outperforms all the others. Finally, they show that the best performances are achieved by fusing the FSNS and the GMM-supervectors systems.

In this paper, we present a system for aircraft automatic target recognition (ATR) using inverse synthetic aperture radar (ISAR) and based on knowledge discovery from data process adapted to radar domain. We propose a method for target shape extraction from ISAR images based on the combination of two methods, SUSAN modified and active deformable contours via level set. In the second part of this work, we propose to fuse two commonly used shape descriptors algorithms based on moments invariant and Fourier descriptors. We have investigated the impact of the information fusion on the recognition rate. The classification scheme is ensured using support vector machine (SVM) classifier. Several combination strategies are compared at score/decision/feature level. Experimental results of the proposed method are provided and discussed.

This paper analyzes the use of visual words, as low-level image features, for learning and categorizing images. We show that this problem can be reduced to a simultaneous weighting of appropriate features and detection of clusters in a binary feature space. A probabilistic model is then proposed to quantify the effectiveness of visual words when treated as binary features. In order to learn the model, we consider a maximum a posteriori (MAP) approach. Experimental results are presented to illustrate the feasibility and merits of our approach.

In this paper, we provide theoretical results on the problem of optimal stream weight selection for the two stream classification problem. It is shown that in the presence of estimation or modeling errors using stream weights can decrease the total classification error. Specifically, we show that stream weights should be selected to be proportional to the feature stream reliability and informativeness. Next, we turn our attention to the problem of unsupervised stream weights computation in real tasks. Based on the theoretical results we propose to use models and ldquoanti-modelsrdquo (class-specific background models) to estimate stream weights. A nonlinear function of the ratio of the inter- to intra-class distance is proposed for stream weight estimation. The resulting unsupervised stream weight estimation algorithm is evaluated on both artificial data and on the problem of audiovisual speech classification. Finally, the proposed algorithm is extended to the problem of audiovisual speech recognition. It is shown that the proposed algorithms achieve results comparable to the supervised minimum-error training approach for classification tasks under most testing conditions.

In this paper, a novel solution to the problem of unsupervised stream weight estimation for multi-stream classification tasks is proposed. Our work is based on theoretical results in A. Potamianos et al. (2006) for the two-class problem were the optimal stream weights are shown to be inversely proportional to the single stream misclassification error. These two-class results are applied to the multi-class problem by using models and "anti-models" (class-specific background models) thus posing the multi-class problem as multiple two-class problems. A nonlinear function of the ratio of the inter- to intra-class distance is proposed as an estimate for single stream classification error and used for stream weight estimation. The proposed unsupervised stream weight estimation algorithm is evaluated on both artificial data and on the problem of audio-visual speech recognition. It is shown that the proposed algorithm achieves results comparable to the supervised minimum-error training approach under most testing conditions.

The generalized linear discriminant sequence (GLDS) kernel has been shown to provide very good performance and efficiency at the NIST Speaker Recognition Evaluations (SRE) in the last few years. This kernel is based on an explicit map of polynomial expansions of input frames which, because of practical limitations, have to be of a degree less or equal to three. In this paper, we consider an extension of the GLDS kernel to allow not only any polynomial degree but also any embedding, including infinite-dimensional ones associated with Mercer kernels (such as Gaussian kernels). It turns out that the resulting kernels belong to the family of posterior covariance kernels. However, their exact ldquokernelizedrdquo form involves the computation of the Gram matrix on background data, and may be intractable when the background corpus is very large (which is the case in speaker verification). To overcome this problem, we use a low-rank approximation of the Gram matrix to provide an approximate but tractable form of these kernels. We then present comparative experiments on NIST SRE 2005. The results show that our sequence kernel outperforms the GLDS one, and gives similar (individual) performances to the traditional universal background model-Gaussiam mixture model (UBM-GMM) system. As expected, the fusion of both improves the scores.

Recently we introduced a new and simple denoising algorithm, called SPACE, that yielded promising preliminary results in noise robust speech recognition. SPACE is essentially based on GMM modeling of clean an noisy speech. In this paper, we evaluate the performance of SPACE on Aurora2 and show that they are globally not satisfactory, essentially because the Gaussian correspondence assumption is not verified. We then propose a new training procedure for the GMMs that achieves a better Gaussian correspondence. We further develop a simple adaptation algorithm to handle unknown environments that preserves the Gaussian correspondence. We evaluate the new denoising algorithm on Aurora2. The results show that it outperforms the multistyle models, sometimes significantly, on the three test sets of Aurora2

In this paper, we provide theoretical results on the problem of optimal stream weight selection for the multi-stream classification problem. It is shown, that in the presence of estimation or modeling errors using stream weights can decrease the total classification error. The stream weights that minimize classification estimation error are shown to be inversely proportional to the single-stream pdf estimation error. It is also shown that under certain conditions, the optimal stream weights are inversely proportional to the single-stream classification error. We apply these results to the problem of audio-visual speech recognition and experimentally verify our claims. The applicability of the results to the problem of unsupervised stream weight estimation is also discussed

In speaker recognition, several recent studies attempt to integrate prior knowledge in order to make better distinction between speakers. This paper studies the relative speaker discriminative power of speech transient and steady zones. An automatic segmentation is used to localize these two types of zones. Experiments are carried out with the KIST 2003 speaker evaluation database. They show that transient frames, in the neighborhood of segment boundaries, are more speaker discriminative than the middle frames of long segments which correspond to the steady parts of phones. In addition, this study leads to a higher efficiency system (than the baseline one) by processing roughly three times less data during the scoring stage.

In all the efficient speaker recognition systems, the decision score is based on the average of the likelihood ratio computed on each frame of the sentence. Except for the non speech frames which are rejected, each one has the same weight in this summation. This paper deals with the study of the speaker relevance of each frame. An automatic segmentation provides quasi stationary segments of variable length; a weight is allocated to each frame in function of its segment position and a weighted mean of the likelihood ratio is then computed. Experiments are performed with NIST 2003 speaker evaluation database. They show that the frames near segment frontiers, that is to say the transient ones, are more speaker relevant than the middle frames of long segments which correspond to the steady parts of the phones.

Colloque avec actes et comité de lecture. nationale.

Colloque avec actes et comité de lecture. internationale.

Colloque avec actes et comité de lecture. internationale.

The multiscale autoregressive (MAR) framework was introduced to support the development of optimal multiscale statistical signal processing. Its power resides in the fast and flexible algorithms to which it leads. While the MAR framework was originally motivated by wavelets, the link between these two worlds has been previously established only in the simple case of the Haar wavelet. The first contribution of this paper is to provide a unification of the MAR framework and all compactly supported wavelets as well as a new view of the multiscale stochastic realization problem. The second contribution of this paper is to develop wavelet-based approximate internal MAR models for stochastic processes. This will be done by incorporating a powerful synthesis algorithm for the detail coefficients which complements the usual wavelet reconstruction algorithm for the scaling coefficients. Taking advantage of the statistical machinery provided by the MAR framework, we will illustrate the application of our models to sample-path generation and estimation from noisy, irregular, and sparse measurements.