Mondher Frikha
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Mondher Frikha has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Kadria Ezzine
ATISP, ENET’COM, Sfax University, Sfax, Tunisia
6 shared publications - Abir Chaari
National School of Electronics and Communication of Sfax, Tunisia
3 shared publications - Girish Gidaye
K. J. Somaiya College of Engineering, Vidyalankar Institute of Technology, Navi Mumbai, India
3 shared publications - Jagannath Nirmal
K J Somaiya School of Engineering, Somaiya Vidyavihar University
3 shared publications - Joseph Di Martino
LORIA—Laboratoire Lorrain de Recherche en Informatique et ses Applications, B.P. 239, 54506 Vandœuvre-lès-Nancy, France
3 shared publications - Hadjer Zebidi
National School of Electronics and Telecommunications of Sfax (ENET’Com), University of Sfax
2 shared publications
Latest Publications
- 1 January 2026
- Published by American Scientific Publishing Group (ASPG) LLC in Journal of Intelligent Systems and Internet of Things
- Vol. 18 (2) , 220-20-238
Assessing Quality Attributes of Microservices in Hadoop and Spark Clusters: A Performance Benchmarking Approach in Dockerized and Non-Dockerized Architectures
American scientific publishing group
- 1 January 2026
- Published by American Scientific Publishing Group (ASPG) LLC in Journal of Intelligent Systems and Internet of Things
- Vol. 18 (2) , 220-20-238
Assessing Quality Attributes of Microservices in Hadoop and Spark Clusters: A Performance Benchmarking Approach in Dockerized and Non-Dockerized Architectures
American scientific publishing group
- 1 December 2025
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 29 (1) , 1
A multilingual speech analysis framework for robust and explainable early detection of Parkinson’s disease
- 1 December 2025
- Published by Springer Nature in International Journal of Speech Technology
- Vol. 29 (1) , 1
A multilingual speech analysis framework for robust and explainable early detection of Parkinson’s disease
- 30 January 2025
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 11 (2) , 025017
Hybrid data augmentation strategies for robust deep learning classification of corneal topographic maptopographic map
Deep learning has emerged as a powerful tool in medical imaging, particularly for corneal topographic map classification. However, the scarcity of labeled data poses a significant challenge to achieving robust performance. This study investigates the impact of various data augmentation strategies on enhancing the performance of a customized convolutional neural network model for corneal topographic map classification. We propose a hybrid data augmentation approach that combines traditional transformations, generative adversarial networks, and specific generative models. Experimental results demonstrate that the hybrid data augmentation method, achieves the highest accuracy of 99.54%, significantly outperforming individual data augmentation techniques. This hybrid approach not only improves model accuracy but also mitigates overfitting issues, making it a promising solution for medical image classification tasks with limited data availability.
- 30 January 2025
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 11 (2) , 025017
Hybrid data augmentation strategies for robust deep learning classification of corneal topographic maptopographic map
Deep learning has emerged as a powerful tool in medical imaging, particularly for corneal topographic map classification. However, the scarcity of labeled data poses a significant challenge to achieving robust performance. This study investigates the impact of various data augmentation strategies on enhancing the performance of a customized convolutional neural network model for corneal topographic map classification. We propose a hybrid data augmentation approach that combines traditional transformations, generative adversarial networks, and specific generative models. Experimental results demonstrate that the hybrid data augmentation method, achieves the highest accuracy of 99.54%, significantly outperforming individual data augmentation techniques. This hybrid approach not only improves model accuracy but also mitigates overfitting issues, making it a promising solution for medical image classification tasks with limited data availability.
- 20 December 2024
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 11 (1) , 015039
Automated feature selection for early keratoconus screening optimization
In this paper, an automated feature selection (FS) method is presented to optimize machine learning (ML) models’ performances, enhancing early keratoconus screening. A total of 448 parameters were analyzed from a dataset comprising 3162 observations sourced from the swept source optical coherence tomography imaging system developed by the Chinese Academy of Sciences Institute of Automation (SS-1000 CASIA OCT) and electronic health records (EHR). To identify the most relevant features, the analysis of variance (ANOVA) method was used in this study. The performance of three classifiers namely K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Artificial Neural Networks (ANN) was evaluated, yielding classification accuracies of 96.79% and 96.68% for KNN, 98.95% and 97.08% for SVM, and 95.64% and 95.62% for ANN when distinguishing between 2 and 4 keratoconus classes, respectively. The results show that selecting 50 features can significantly improve the performance of the model while reducing the computation time. The automated feature selection method can also assist ophthalmologists in better understanding the links between various ocular characteristics and keratoconus, potentially leading to advances in early diagnosis, risk prediction, and clinical management of this condition.
- 20 December 2024
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 11 (1) , 015039
Automated feature selection for early keratoconus screening optimization
In this paper, an automated feature selection (FS) method is presented to optimize machine learning (ML) models’ performances, enhancing early keratoconus screening. A total of 448 parameters were analyzed from a dataset comprising 3162 observations sourced from the swept source optical coherence tomography imaging system developed by the Chinese Academy of Sciences Institute of Automation (SS-1000 CASIA OCT) and electronic health records (EHR). To identify the most relevant features, the analysis of variance (ANOVA) method was used in this study. The performance of three classifiers namely K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Artificial Neural Networks (ANN) was evaluated, yielding classification accuracies of 96.79% and 96.68% for KNN, 98.95% and 97.08% for SVM, and 95.64% and 95.62% for ANN when distinguishing between 2 and 4 keratoconus classes, respectively. The results show that selecting 50 features can significantly improve the performance of the model while reducing the computation time. The automated feature selection method can also assist ophthalmologists in better understanding the links between various ocular characteristics and keratoconus, potentially leading to advances in early diagnosis, risk prediction, and clinical management of this condition.
- Open Access
- 1 October 2024
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 13 (5) , 3568-3578
Handwritten Arabic words detection using Faster R-CNN in IFN/ENIT dataset
Recognizing Arabic offline handwritten words still faces various challenges because of the diversity of writing styles and the overlap between the words and characters. Therefore, building an effective system to solve these challenges has always been difficult, which has led to a lack of published research in this field. This study introduces two new models to recognize handwritten Arabic words based on the Faster region-convolution neural network (Faster R-CNN). These models employ two pre-trained networks during the feature extraction phase: The visual geometry group-16 (VGG-16) network and the residual network (ResNet50) network. To help with overlapping detections and make localization more accurate, a soft non-maximum suppression (Soft-NMS) strategy is used in post-processing. Models are independently trained and tested on two groups of data from the Institut Für Nachrichtentechnik/Ecole Nationale d’Ingénieurs de Tunis (IFN/ENIT) dataset. The first group includes one word in each image, while the second contains multiple words. Test results showed that the proposed models give excellent results compared to others. The results of VGG16 and ResNet50 with the first dataset reached accuracy rates of 100% and 99.5%, respectively. Meanwhile, the accuracy of the second group reached 91.4% and 100% with VGG16 and ResNet50, respectively.
- 1 October 2024
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 13 (5) , 3568-3578
Open AccessHandwritten Arabic words detection using Faster R-CNN in IFN/ENIT dataset
Recognizing Arabic offline handwritten words still faces various challenges because of the diversity of writing styles and the overlap between the words and characters. Therefore, building an effective system to solve these challenges has always been difficult, which has led to a lack of published research in this field. This study introduces two new models to recognize handwritten Arabic words based on the Faster region-convolution neural network (Faster R-CNN). These models employ two pre-trained networks during the feature extraction phase: The visual geometry group-16 (VGG-16) network and the residual network (ResNet50) network. To help with overlapping detections and make localization more accurate, a soft non-maximum suppression (Soft-NMS) strategy is used in post-processing. Models are independently trained and tested on two groups of data from the Institut Für Nachrichtentechnik/Ecole Nationale d’Ingénieurs de Tunis (IFN/ENIT) dataset. The first group includes one word in each image, while the second contains multiple words. Test results showed that the proposed models give excellent results compared to others. The results of VGG16 and ResNet50 with the first dataset reached accuracy rates of 100% and 99.5%, respectively. Meanwhile, the accuracy of the second group reached 91.4% and 100% with VGG16 and ResNet50, respectively.
- Open Access
- 2 August 2024
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 18 (14) , 103-119
Masked Face Recognition Using Bag of CNN: Robust Local Feature Extraction and Region of Interest
Face recognition remains a crucial issue in computer vision with various applications. This paper introduces an adapted method to tackle the challenges posed by mask-wearing during the COVID-19 pandemic. We propose modifications to the bag of convolutional neural networks (BoCNN) framework, which combines CNNs and the bag of words (BoW) approach. Our main contribution is customizing the BoCNN algorithm to identify faces with masks by focusing on the visible facial regions, particularly the eyes and eyebrows. Facial landmarks are detected, and the region of interest (ROI) is extracted using techniques such as Media Pipe. A pre-trained CNN is then applied to sections within the ROI, enabling robust feature extraction that captures intricate details such as lighting variations and facial expressions while reducing the impact of mask occlusions. The extracted features are pooled to create a comprehensive representation for recognition. Extensive experiments on the labeled faces in the wild (LFW) dataset, including masked face images, demonstrate the superiority of our adapted BoCNN approach over traditional BoW and deep feature extraction methods, especially accurately recognizing masked faces. Additionally, we assess the generalizability of our method across multiple datasets and discuss potential limitations and future research directions. The proposed BoCNN-based solution proves effective in recognizing masked faces, making it highly relevant for applications in security, human-computer interaction, and various other domains affected by the COVID-19 pandemic.
- 2 August 2024
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 18 (14) , 103-119
Open AccessMasked Face Recognition Using Bag of CNN: Robust Local Feature Extraction and Region of Interest
Face recognition remains a crucial issue in computer vision with various applications. This paper introduces an adapted method to tackle the challenges posed by mask-wearing during the COVID-19 pandemic. We propose modifications to the bag of convolutional neural networks (BoCNN) framework, which combines CNNs and the bag of words (BoW) approach. Our main contribution is customizing the BoCNN algorithm to identify faces with masks by focusing on the visible facial regions, particularly the eyes and eyebrows. Facial landmarks are detected, and the region of interest (ROI) is extracted using techniques such as Media Pipe. A pre-trained CNN is then applied to sections within the ROI, enabling robust feature extraction that captures intricate details such as lighting variations and facial expressions while reducing the impact of mask occlusions. The extracted features are pooled to create a comprehensive representation for recognition. Extensive experiments on the labeled faces in the wild (LFW) dataset, including masked face images, demonstrate the superiority of our adapted BoCNN approach over traditional BoW and deep feature extraction methods, especially accurately recognizing masked faces. Additionally, we assess the generalizability of our method across multiple datasets and discuss potential limitations and future research directions. The proposed BoCNN-based solution proves effective in recognizing masked faces, making it highly relevant for applications in security, human-computer interaction, and various other domains affected by the COVID-19 pandemic.
- Open Access
- 9 February 2024
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 18 (03) , 88-103
Convolutional Neural Network Architectures for Gender, Emotional Detection from Speech and Speaker Diarization
This paper introduces three system architectures for speaker identification that aim to overcome the limitations of diarization and voice-based biometric systems. Diarization systems utilize unsupervised algorithms to segment audio data based on the time boundaries of utterances, but they do not distinguish individual speakers. On the other hand, voice-based biometric systems can only identify individuals in recordings with a single speaker. Identifying speakers in recordings of natural conversations can be challenging, especially when emotional shifts can alter voice characteristics, making gender identification difficult. To address this issue, the proposed architectures include techniques for gender, emotion, and diarization at either the segment or group level. The evaluation of these architectures utilized two speech databases, namely VoxCeleb and RAVDESS (Ryerson audio-visual database of emotional speech and song) datasets. The findings reveal that the proposed approach outperforms the strategy level in terms of recognition results, despite the real-time processing advantage of the latter. The challenge of identifying multiple speakers engaging in a conversation while considering emotional changes that impact speech is effectively addressed by the proposed architectures. The data indicates that the gender and emotion classification of diarization achieves an accuracy of over 98 percent. These results suggest that the proposed speech-based approach can achieve highly accurate speaker identification.
- 9 February 2024
- Published by International Association of Online Engineering (IAOE) in International Journal of Interactive Mobile Technologies (iJIM)
- Vol. 18 (03) , 88-103
Open AccessConvolutional Neural Network Architectures for Gender, Emotional Detection from Speech and Speaker Diarization
This paper introduces three system architectures for speaker identification that aim to overcome the limitations of diarization and voice-based biometric systems. Diarization systems utilize unsupervised algorithms to segment audio data based on the time boundaries of utterances, but they do not distinguish individual speakers. On the other hand, voice-based biometric systems can only identify individuals in recordings with a single speaker. Identifying speakers in recordings of natural conversations can be challenging, especially when emotional shifts can alter voice characteristics, making gender identification difficult. To address this issue, the proposed architectures include techniques for gender, emotion, and diarization at either the segment or group level. The evaluation of these architectures utilized two speech databases, namely VoxCeleb and RAVDESS (Ryerson audio-visual database of emotional speech and song) datasets. The findings reveal that the proposed approach outperforms the strategy level in terms of recognition results, despite the real-time processing advantage of the latter. The challenge of identifying multiple speakers engaging in a conversation while considering emotional changes that impact speech is effectively addressed by the proposed architectures. The data indicates that the gender and emotion classification of diarization achieves an accuracy of over 98 percent. These results suggest that the proposed speech-based approach can achieve highly accurate speaker identification.
- Open Access
- 2 November 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (21) , 11988
Any-to-One Non-Parallel Voice Conversion System Using an Autoregressive Conversion Model and LPCNet Vocoder
We present an any-to-one voice conversion (VC) system, using an autoregressive model and LPCNet vocoder, aimed at enhancing the converted speech in terms of naturalness, intelligibility, and speaker similarity. As the name implies, non-parallel any-to-one voice conversion does not require paired source and target speeches and can be employed for arbitrary speech conversion tasks. Recent advancements in neural-based vocoders, such as WaveNet, have improved the efficiency of speech synthesis. However, in practice, we find that the trajectory of some generated waveforms is not consistently smooth, leading to occasional voice errors. To address this issue, we propose to use an autoregressive (AR) conversion model along with the high-fidelity LPCNet vocoder. This combination not only solves the problems of waveform fluidity but also produces more natural and clear speech, with the added capability of real-time speech generation. To precisely represent the linguistic content of a given utterance, we use speaker-independent PPG features (SI-PPG) computed from an automatic speech recognition (ASR) model trained on a multi-speaker corpus. Next, a conversion model maps the SI-PPG to the acoustic representations used as input features for the LPCNet. The proposed autoregressive structure enables our system to produce the following prediction step outputs from the acoustic features predicted in the previous step. We evaluate the effectiveness of our system by performing any-to-one conversion pairs between native English speakers. Experimental results show that the proposed method outperforms state-of-the-art systems, producing higher speech quality and greater speaker similarity.
- 2 November 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (21) , 11988
Open AccessAny-to-One Non-Parallel Voice Conversion System Using an Autoregressive Conversion Model and LPCNet Vocoder
We present an any-to-one voice conversion (VC) system, using an autoregressive model and LPCNet vocoder, aimed at enhancing the converted speech in terms of naturalness, intelligibility, and speaker similarity. As the name implies, non-parallel any-to-one voice conversion does not require paired source and target speeches and can be employed for arbitrary speech conversion tasks. Recent advancements in neural-based vocoders, such as WaveNet, have improved the efficiency of speech synthesis. However, in practice, we find that the trajectory of some generated waveforms is not consistently smooth, leading to occasional voice errors. To address this issue, we propose to use an autoregressive (AR) conversion model along with the high-fidelity LPCNet vocoder. This combination not only solves the problems of waveform fluidity but also produces more natural and clear speech, with the added capability of real-time speech generation. To precisely represent the linguistic content of a given utterance, we use speaker-independent PPG features (SI-PPG) computed from an automatic speech recognition (ASR) model trained on a multi-speaker corpus. Next, a conversion model maps the SI-PPG to the acoustic representations used as input features for the LPCNet. The proposed autoregressive structure enables our system to produce the following prediction step outputs from the acoustic features predicted in the previous step. We evaluate the effectiveness of our system by performing any-to-one conversion pairs between native English speakers. Experimental results show that the proposed method outperforms state-of-the-art systems, producing higher speech quality and greater speaker similarity.
- Open Access
- 2 November 2023
- Published by MDPI AG in Preprints.org
Any-to-One Non-Parallel Voice Conversion System Using an Autoregressive Conversion Model and LPCNet Vocoder
We present an any-to-one voice conversion (VC) system, using an autoregressive model and LPCNet vocoder, aimed to enhance the converted speech in terms of naturalness, intelligibility, and speaker similarity. As the name implies, non-parallel any-to-one voice conversion does not require paired source and target speeches and can be employed for arbitrary speech conversion tasks. Recent advancements in neural-based vocoders, such as WaveNet, have improved the efficiency of speech synthesis. However, in practice, we find that the trajectory of some generated waveforms is not consistently smooth, leading to occasional voice errors. To address this issue, we propose to use an autoregressive (AR) conversion model along with the high-fidelity LPCNet vocoder. This combination not only solves the problems of waveform fluidity but also produces more natural and clear speech, with the added capability of real-time speech generation. To precisely represent the linguistic content of a given utterance, we use speaker-independent PPG features (SI-PPG) computed from an automatic speech recognition (ASR) model trained on a multi-speaker corpus. Next, a conversion model maps the SI-PPG to the acoustic representations used as input features for the LPCNet. The proposed autoregressive structure enables our system to produce the following prediction step outputs from the acoustic features predicted in the previous step. We evaluate the effectiveness of our system by performing any-to-one conversion pairs between native English speakers. Experimental results show that the proposed method outperforms state-of-the-art systems, producing higher speech quality and greater speaker similarity.
- 2 November 2023
- Published by MDPI AG in Preprints.org
Open AccessAny-to-One Non-Parallel Voice Conversion System Using an Autoregressive Conversion Model and LPCNet Vocoder
We present an any-to-one voice conversion (VC) system, using an autoregressive model and LPCNet vocoder, aimed to enhance the converted speech in terms of naturalness, intelligibility, and speaker similarity. As the name implies, non-parallel any-to-one voice conversion does not require paired source and target speeches and can be employed for arbitrary speech conversion tasks. Recent advancements in neural-based vocoders, such as WaveNet, have improved the efficiency of speech synthesis. However, in practice, we find that the trajectory of some generated waveforms is not consistently smooth, leading to occasional voice errors. To address this issue, we propose to use an autoregressive (AR) conversion model along with the high-fidelity LPCNet vocoder. This combination not only solves the problems of waveform fluidity but also produces more natural and clear speech, with the added capability of real-time speech generation. To precisely represent the linguistic content of a given utterance, we use speaker-independent PPG features (SI-PPG) computed from an automatic speech recognition (ASR) model trained on a multi-speaker corpus. Next, a conversion model maps the SI-PPG to the acoustic representations used as input features for the LPCNet. The proposed autoregressive structure enables our system to produce the following prediction step outputs from the acoustic features predicted in the previous step. We evaluate the effectiveness of our system by performing any-to-one conversion pairs between native English speakers. Experimental results show that the proposed method outperforms state-of-the-art systems, producing higher speech quality and greater speaker similarity.