Zied Lachiri

Emotion recognition from physiological signals using fusion of wavelet based features

In this paper we propose a new system for human emotion recognition based on multi resolution analysis of physiological signals. In our study we have used four kinds of bio signals EMG, RESP, ECG and SC recorded at the University of Augsburg. Daubechies Symlet, Haar and Morlet wavelet transform were applied to analyze the non-stationary signals. Physiological features was extracted from the most relevant wavelet coefficients and the feature vectors obtained from each signal were combined using multimodal fusion technique to construct one feature vector for each emotion. A support vector machine (SVM) was adopted as a pattern classifier, an improved recognition accuracy of 95% was obtained and it clearly proves the performance of our new wavelet based approach in emotion recognition.

Multi-class SVM for stressed speech recognition

This paper deals with a new automatic stressed recognition system based on kernel classification. We extracted advanced acoustic features from the stressed signals and employed a multi-class Support Vector Machines with different kernels to recognize speech utterances under stress. Gammatone Frequency Cepstral Coefficients are also established. The system implemented is tested using isolated words from SUSAS database with 4 classes: Neutral, Angry, Lombard and Loud. Experimental results show that the best performance is obtained when we use the auditory feature with different descriptors combination but it depends on the type of the kernel used.

Wavelet analysis of frequency chaos game signal: a time-frequency signature of the C. elegans DNA

Challenging tasks are encountered in the field of bioinformatics. The choice of the genomic sequence’s mapping technique is one the most fastidious tasks. It shows that a judicious choice would serve in examining periodic patterns distribution that concord with the underlying structure of genomes. Despite that, searching for a coding technique that can highlight all the information contained in the DNA has not yet attracted the attention it deserves. In this paper, we propose a new mapping technique based on the chaos game theory that we call the frequency chaos game signal (FCGS). The particularity of the FCGS coding resides in exploiting the statistical properties of the genomic sequence itself. This may reflect important structural and organizational features of DNA. To prove the usefulness of the FCGS approach in the detection of different local periodic patterns, we use the wavelet analysis because it provides access to information that can be obscured by other time-frequency methods such as the Fourier analysis. Thus, we apply the continuous wavelet transform (CWT) with the complex Morlet wavelet as a mother wavelet function. Scalograms that relate to the organism Caenorhabditis elegans (C. elegans) exhibit a multitude of periodic organization of specific DNA sequences.

Gabor Filterbank Features for Robust Speech Recognition

Several research studies have shown that the robustness and performance of speech recognition systems can be improved using physiologically inspired filterbank based on Gabor filters. In this paper, we proposed a feature extraction method based on 59 two-dimensional Gabor filterbank. The use of these set of filters aims to extracting specific modulation frequencies and limiting the redundancy on feature level. The recognition performance of our feature extraction method is evaluated in isolated words extracted from TIMIT corpus. The obtained results demonstrate that the proposed extraction method gives better recognition rates to those obtained using the classic methods MFCC, PLP and LPC.

Classification of Helitron’s Types in the C.elegans Genome based on Features Extracted from Wavelet Transform and SVM Methods

Helitrons, a sub-class of the Transposable elements class 2, are considered as an important DNA type. In fact, they contribute in mechanism’s evolution. Till now, these elements are not well studied using the automatic tools. In fact, the researches done in helitrons recognition are based only on biological experiments. In this paper, we propose an automatic method for characterizing helitrons by global signature and classifying the helitron’s types in C.elegans genome. For this goal, we used the Complex Morlet Wavelet Transform to generate helitron’s signatures (helitron’s scalograms presentation) and to extract the features of each category. Then, we used the SVM-classifier to classify these 10 helitron’s families. After testing different kernels and using the cross validation function, we present the best classification results given by the RBF-kernel with c=60, σ=0. 0000000015625 and OAO approach.

Landmine detection improvement using one-class SVM for unbalanced data

Ground Penetrating Radar (GPR) has been a precious tool for humanitarian demining. The GPR scans the ground and delivers a three-dimensional matrix representing three types of data: Ascan, Bscan and Cscan. The Ascan data represents the response from a reflection signal of a pulse emitted by the GPR at a given position. In the proposed landmine detection method, the Ascan data is normalized and then classified using Kernel based One Class Support Vector Machine (OSVM). In fact, OSVM has the main advantage of handling unbalanced data, where is not the case for multiclass SVM. Our landmine detection method was tested and evaluated on the MACADAM database which is composed of 11 scenarios of landmines and 3 scenarios of inoffensive objects (wood stick, SodaCan, pine, stone). Experimental results have shown the superiority of the RBF kernel OSVM over others kernel functions based multiclass SVM in term of classification accuracy especially, as landmine data is unbalanced.

Emotion recognition system based on physiological signals with Raspberry Pi III implementation

Human machine interaction fieldhas potentialapplications in different domainssuch as medicine therapies for vulnerable persons. Thus, allowing the machine to identify and understand emotional states is one of the primordial stages for affective interactivity with Humans. Recent studies have proved that physiological signals contribute to recognize the emotion. In this paper, we aim to classify the affective states into two defined classes in arousal-valence model using peripheral physiological signals. For this aim, we explored the recent multimodal MAHNOB-HCI database that contains the bodily responses of 24 participants to 20 affective videos. After preprocessing the data and extracting features, we classified the emotion using the Support Vector Machine (SVM). The classification stage was implemented on Raspberry Pi III model B using Python platform. The obtained results are encouraging compared to recent related works.

Emotional speaker recognition in real life conditions using multiple descriptors and i-vector speaker modeling technique

Emotional speaker recognition under real life conditions becomes an urgent need for several applications. This paper proposes a novel approach using multiple feature extraction methods and i-vector modeling technique in order to improve emotional speaker recognition under real conditions. The performance of the proposed approach is evaluated on real condition speech signal (IEMOCAP corpus) under clean and noisy environments using various SNR levels. We examined divers known spectral features in speaker recognition (MFCC, LPCC and RASTA-PLP) and performed combined features called MFCC-SDC coefficients. The feature vectors are then classified using the multiclass Support Vector Machines (SVM). Experimental results illustrate good robustness of the proposed system against talking conditions (emotions) and against real life environment (noise). Besides, results reveal that MFCC-SDC features outperforms the conventional MFCCs.

Classification of speech under stress based on cepstral features and one-class SVM

This paper presents an approach that aims to recognize stressed speech utterances. Our work consists of extracting features using Mel Frequency Cepstral Coefficients (MFCC) and Gammatone Frequency Cepstral Coefficients (GFCC). Indeed, these features are classified with One-class Support Vector Machines (OC-SVM). The results of the proposed method are obtained by conducting speech samples of four stressed states from the SUSAS database. The system provides good performances with accuracy rate exceeding 98% with the different features extracted from the stressed database. A comparison between the classification accuracies obtained with OC-SVM and those given when we apply other multiclass Support Vectors Machines approaches is also presented.

Emotion sensing from physiological signals using three defined areas in arousal-valence model

This paper aims to recognize the human emotional states into three defined areas in arousal-valence evaluation: Corresponding to calm, medium aroused, and excited, unpleasant, neutral valence and pleasant. And thanks to the relevance of the peripheral physiological signals in emotion recognition issue, we used in our contribution the multimodal dataset MAHNOB-HCI. In this database, there are emotional bodily responses of twenty four participants after watching twenty affective stimuli videos. In our work, we focused on the ElectroCardioGram (ECG), Galvanic Skin Response (GSR), Skin Temperature (Temp) and Respiration Volume (RESP). To accomplish our purpose, we pre-process the data, extract 169 features and finally, classify the emotional states by using the support vector machine (SVM). As a first step, we classify each signal to know the most relevant physiological signal for emotion assessing, then a level feature fusion is applied to compare our approach to related work. According to previous studies, our obtained results are promising and show that the respiration and electrocardiogram are the most relevant.