We present an any-to-one voice conversion (VC) system, using an autoregressive model and LPCNet v... more 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 utte...
TELKOMNIKA (Telecommunication Computing Electronics and Control)
Recommendation systems (RS) are crucial for social networking sites. Without it, finding precise ... more Recommendation systems (RS) are crucial for social networking sites. Without it, finding precise products is harder. However, existing systems lack adequate efficiency, especially with big data. This paper presents a prototype cloud-based recommendation system for processing big data. The proposed work is implemented by utilizing the matrix factorization method with three approaches. In the first approach, singular value decomposition (SVD) is used, which is an old and traditional recommendation technique. The second recommendation approach is fine-tuned using the alternating least squares (ALS) algorithm with Apache Spark. Finally, the deep neural network (DNN) algorithm is utilized with TensorFlow. This study solves the challenge of handling large-scale datasets in the collaborative filtering (CF) technique after tuning the algorithms by adjusting the parameters in the second approach, which uses machine learning, as well as in the third approach, which uses deep learning. Furthermore, the results of these two approaches outperformed conventional techniques and achieved an acceptable computational time. The dataset size is about 1.5 GB and it is collected from the Goodreads website API. Moreover, the Hadoop distributed file system (HDFS) is used as cloud storage instead of the computer's local disk for handling larger dataset sizes in the future.
HAL (Le Centre pour la Communication Scientifique Directe), May 27, 2021
HAL is a multidisciplinary open access archive for the deposit and dissemination of scientific re... more HAL is a multidisciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
2016 IEEE/ACS 13th International Conference of Computer Systems and Applications (AICCSA), 2016
Thresholding is a popular image segmentation method that converts gray-level image into binary im... more Thresholding is a popular image segmentation method that converts gray-level image into binary image. The problem of thresholding has been quite extensively studied for many years in order to get an optimum threshold value. The multi-level thresholding becomes very computationally challenges. In this paper, a novel multilevel thresholding method based on particle swarm optimization (PSO) algorithm is proposed, or it seems to be the best tool, to maximize the Kapur and Otsu objective functions. We employed the properties of discriminate analysis using Kapur and Otsu methods to render the optimal thresholding technics more applicable and effective. The obtained result and the comparative study illustrate the algorithm's outstanding performances in segmenting both the grey level image and the MRI scans.
Measurement of transients of ambient power-line frequency magnetic fields is becoming increasingl... more Measurement of transients of ambient power-line frequency magnetic fields is becoming increasingly important because of the possibility that exposure to such fields may produce effects to communication systems and various electronic devices as well as human health. Many studies have been concerned with magnetic field measurements over a wide frequency range but small attention has been paid to the measurement of transient fields. The objective of this study is to investigate the feasibility of developing a magnetic field probe operating in a broad frequency range, including the power line frequency (60 Hz), which would also be suitable for measurements of transient fields (their magnitude and phase). The sensor was modeled, simulated using computer simulation and finally designed. Its performance was tested in time and frequency domains. The sensor, which provides constant amplitude and phase response from 30 Hz to around 5 MHz for the magnetic field, has a minimal response to the electric field (30 dB or less), a sensitivity of 0.48 V/A $m\sp{-1}$ and wide dynamic range (greater than 40 dB).
Recent advances in the field of speaker recognition have proved to highly outperform algorithms. ... more Recent advances in the field of speaker recognition have proved to highly outperform algorithms. However this performance degrades when limited data are presented. This paper presents examples on how Support Vector Machines (SVM) can improve speaker recognition for short utterance data duration. The main contribution in this approach is the use of new vectors when training and testing data are limited. We show how different kernels function of SVM can be used to validate the new approach with different speakers from different databases.
This paper discusses the accurate measurement of formant frequencies using Cesptral and LPC metho... more This paper discusses the accurate measurement of formant frequencies using Cesptral and LPC method. Each algorithm was implemented with Matlab and was applied in the aim to evaluate the precision of both designed techniques. The conceived Cepstral algorithm is a frequency method based on picking peaks from the Cepstrally-smoothed frequency spectrum of the speech signal. Cepstral smoothing is a nonparametric method that attempts to remove the effect of glottal pulsing to obtain the spectral envelope corresponding to the vocal tract response. The obtained result, i.e. the Cepstrum, was then used to estimate the smoothed spectrum. Formant frequencies are estimated from the smoothed speech spectrum by adding constraints on the formant frequency ranges. The four highest peaks are typically classified as the first four formants. However, the LPC algorithm estimate formant frequencies from the all pole model of the vocal tract transfer function. The approach relies on the source – filter m...
In this paper, two speech enhancement algorithms (SEAs) based on spectral subtraction (SS) princi... more In this paper, two speech enhancement algorithms (SEAs) based on spectral subtraction (SS) principle have been evaluated for bilateral cochlear implant (BCI) users. Specifically, dual-channel noise power spectral estimation algorithm using power spectral densities (PSD) and cross power spectral density (CPSD) of the observed signals was studied. The enhanced speech signals were obtained using either Dual Channel Non Linear Spectral Subtraction 'DC-NLSS' or Dual-Channel Multi-Band Spectral Subtraction 'DC-MBSS' algorithms. For performance evaluation, some objective speech assessment tests relying on Perceptual Evaluation of Speech Quality (PESQ) score and speech Itakura-Saito (IS) distortion measurement were performed to fix the optimal number of frequency band needed in DC-MBSS algorithm. In order to evaluate the speech intelligibility, subjective listening tests were assessed with 50 normal hearing listeners using a specific BCI simulator and with three deafened BCI patients. Experimental results, obtained using French Lafon database corrupted by an additive babble noise at different Signal-to-Noise Ratios (SNR), showed that DC-MBSS algorithm improves speech understanding better than DC-NLSS algorithm for single and multiple interfering noise sources.
2020 IEEE 4th International Conference on Image Processing, Applications and Systems (IPAS), 2020
Handwritten recognition has received considerable attention in the domain of pattern recognition,... more Handwritten recognition has received considerable attention in the domain of pattern recognition, image processing, over the last few decades. As a consequence of this research effort, several algorithms were developed using different techniques. Particularly, Deep Learning has shown a remarkable capability to handle handwritten recognition in very recent years. The well-known Deep learning techniques are the Convolutional Neuronal Networks (CNNs) and Recurrent Neuronal Networks (RNNs). This paper provides a survey of the most recent handwritten recognition systems. Thus, we present the most significant algorithms for handwritten character/word/text recognition by explaining the different approaches used in the recognition process and we compare them in terms of accuracy.
Knowledge Science, Engineering and Management, 2019
Speaker recognition systems proposed in the literature are conditioned by many factors such as th... more Speaker recognition systems proposed in the literature are conditioned by many factors such as the features used, the type of data to be processed, data duration and the approach to be used. This work presents a novel speaker recognition application using a new dimensional cepstral feature vector for building Gaussian Mixture Models (GMM) for speaker identification and verification systems. Recognition results obtained with the proposed system are compared to the state of the art systems requiring high dimensional feature vectors. We experimentally evaluate this work with speakers taken from TIMIT database. The new system gave substantial improvements even when trained with limited data, which explain the effectiveness of the new feature vectors with this approach which can be comparable to the use additional algorithms or approaches.
2020 International Wireless Communications and Mobile Computing (IWCMC), 2020
An important amount of speech is typically required for speaker identification system development... more An important amount of speech is typically required for speaker identification system development and evaluation. Nowadays, robust speaker identification systems when short utterances are used remains a key consideration for automatic speaker recognition, since a lot of real world applications are able to deal with only limited duration speech data. This paper presents a new approach based on a low complexity solution based on a new feature vectors to build Gaussian Mixture Models (GMM) for speaker identification systems especially when training and testing utterance lengths are reduced. We compared our proposed system to the state-of-the-art based system in Speaker identification. Experiments on TIMIT database were conducted to demonstrate that this new feature vector can outperform the standard GMM-based system and show that there is no need for extra-data to identify the speakers.
2021 International Wireless Communications and Mobile Computing (IWCMC), 2021
This paper presents a new text independent speaker recognition system based on new cepstral Coeff... more This paper presents a new text independent speaker recognition system based on new cepstral Coefficients and i-vector based on probabilistic linear discriminant analysis. This system is designed to be used as a biometric authentication system based on the speaker voice. The experiments were performed on speech data taken from a large number of users and consist of 630 speakers from TIMIT database for different training conditions. The efficiency of the proposed system is observed with the increase of Speaker identification Rate. Experimental results show that the system gives higher recognition performance for different numbers of speakers.
2017 International Conference on Advanced Technologies for Signal and Image Processing (ATSIP), 2017
Speaker identity, the sound of a person's voice, is one of the most important characteristics... more Speaker identity, the sound of a person's voice, is one of the most important characteristics in human communication. Voice conversion (VC) is an emergent problem in voice and speech processing that deals with the process of modifying a speaker's identity. More particularly, the speech signal spoken by the source speaker is modified to sound a sifit had been pronounced by another speaker, referred to as the target speaker. A variety of VC techniques has been proposed since the first appearance of the voice conversion problem. The choice among those techniques represents a compromise between the similarity of the converted voice to the target voice and the quality of the output speech signal, both rated by the used technique. In this paper, we review a comprehensive state-of-the-art of voice conversion techniques while pointing out their advantages and disadvantages. These techniques will be applied in significant and most versatile areas of speech technology; applications that are far beyond speech synthesis.
Acoustic analysis of speech signal enables automatic detection and classification of voice disord... more Acoustic analysis of speech signal enables automatic detection and classification of voice disorders along with its severity. This automatic assessment provides help to the clinician in initial diagnosis of pathological larynx in non-intrusive way. Voice pathologies damage the vocal cords and consequently alter the dynamics (fluctuation speed) of vocal cords. In this article, we have estimated glottal volume velocity waveform (GVVW) from the speech pressure waveforms of healthy and pathological subjects using quasi closed phase (QCP) glottal inverse filtering algorithm to capture altered dynamics of vocal cords. Closed-phase methods revealed notable stability in diverse voice qualities and sub-glottal pressures. The GVVW is the source of significant acoustical clues rooted in speech. The estimated GVVW is then parameterized by various time based, frequency based and Liljencrants-Fant (LF) model based glottal descriptors. Glottal descriptor's vectors have been passed on to stochastic gradient descent (SGD) classifier for voice disorder evaluation. The normal pitch utterance of sustained vowel /a/ quarried from German, English, Arabic and Spanish voice databases is used. Information gain (IG) feature scoring technique is employed to select optimal descriptors and to rank them. Several intra and cross-database experiments were performed to explore the usefulness of glottal descriptors for voice disorder detection, severity detection and classification. Student's t-tests were performed to validate the obtained results.
Recent advances in the speaker recognition (SR) field showed remarkably accurate and outperformin... more Recent advances in the speaker recognition (SR) field showed remarkably accurate and outperforming algorithms. However, their performances drastically degrade when the sparse amount of data is available. Nowadays, recognising a speaker identity when only a small amount of speech data is involved for testing and training remains a key consideration since many real world applications often have access to only speech data having a limited duration. In this study, the authors present a new improved approach, based on new information detected from the speech signal, to improve the task of automatic speaker identification. In doing so, they highlight how the detection of the speaker dialect can be explored to address the research problem related to short utterance SR. Results obtained with the new regional system are presented which provide a comparison between this system and the state-of-the-art systems for speaker identification task.
We present an any-to-one voice conversion (VC) system, using an autoregressive model and LPCNet v... more 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 utte...
TELKOMNIKA (Telecommunication Computing Electronics and Control)
Recommendation systems (RS) are crucial for social networking sites. Without it, finding precise ... more Recommendation systems (RS) are crucial for social networking sites. Without it, finding precise products is harder. However, existing systems lack adequate efficiency, especially with big data. This paper presents a prototype cloud-based recommendation system for processing big data. The proposed work is implemented by utilizing the matrix factorization method with three approaches. In the first approach, singular value decomposition (SVD) is used, which is an old and traditional recommendation technique. The second recommendation approach is fine-tuned using the alternating least squares (ALS) algorithm with Apache Spark. Finally, the deep neural network (DNN) algorithm is utilized with TensorFlow. This study solves the challenge of handling large-scale datasets in the collaborative filtering (CF) technique after tuning the algorithms by adjusting the parameters in the second approach, which uses machine learning, as well as in the third approach, which uses deep learning. Furthermore, the results of these two approaches outperformed conventional techniques and achieved an acceptable computational time. The dataset size is about 1.5 GB and it is collected from the Goodreads website API. Moreover, the Hadoop distributed file system (HDFS) is used as cloud storage instead of the computer's local disk for handling larger dataset sizes in the future.
HAL (Le Centre pour la Communication Scientifique Directe), May 27, 2021
HAL is a multidisciplinary open access archive for the deposit and dissemination of scientific re... more HAL is a multidisciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
2016 IEEE/ACS 13th International Conference of Computer Systems and Applications (AICCSA), 2016
Thresholding is a popular image segmentation method that converts gray-level image into binary im... more Thresholding is a popular image segmentation method that converts gray-level image into binary image. The problem of thresholding has been quite extensively studied for many years in order to get an optimum threshold value. The multi-level thresholding becomes very computationally challenges. In this paper, a novel multilevel thresholding method based on particle swarm optimization (PSO) algorithm is proposed, or it seems to be the best tool, to maximize the Kapur and Otsu objective functions. We employed the properties of discriminate analysis using Kapur and Otsu methods to render the optimal thresholding technics more applicable and effective. The obtained result and the comparative study illustrate the algorithm's outstanding performances in segmenting both the grey level image and the MRI scans.
Measurement of transients of ambient power-line frequency magnetic fields is becoming increasingl... more Measurement of transients of ambient power-line frequency magnetic fields is becoming increasingly important because of the possibility that exposure to such fields may produce effects to communication systems and various electronic devices as well as human health. Many studies have been concerned with magnetic field measurements over a wide frequency range but small attention has been paid to the measurement of transient fields. The objective of this study is to investigate the feasibility of developing a magnetic field probe operating in a broad frequency range, including the power line frequency (60 Hz), which would also be suitable for measurements of transient fields (their magnitude and phase). The sensor was modeled, simulated using computer simulation and finally designed. Its performance was tested in time and frequency domains. The sensor, which provides constant amplitude and phase response from 30 Hz to around 5 MHz for the magnetic field, has a minimal response to the electric field (30 dB or less), a sensitivity of 0.48 V/A $m\sp{-1}$ and wide dynamic range (greater than 40 dB).
Recent advances in the field of speaker recognition have proved to highly outperform algorithms. ... more Recent advances in the field of speaker recognition have proved to highly outperform algorithms. However this performance degrades when limited data are presented. This paper presents examples on how Support Vector Machines (SVM) can improve speaker recognition for short utterance data duration. The main contribution in this approach is the use of new vectors when training and testing data are limited. We show how different kernels function of SVM can be used to validate the new approach with different speakers from different databases.
This paper discusses the accurate measurement of formant frequencies using Cesptral and LPC metho... more This paper discusses the accurate measurement of formant frequencies using Cesptral and LPC method. Each algorithm was implemented with Matlab and was applied in the aim to evaluate the precision of both designed techniques. The conceived Cepstral algorithm is a frequency method based on picking peaks from the Cepstrally-smoothed frequency spectrum of the speech signal. Cepstral smoothing is a nonparametric method that attempts to remove the effect of glottal pulsing to obtain the spectral envelope corresponding to the vocal tract response. The obtained result, i.e. the Cepstrum, was then used to estimate the smoothed spectrum. Formant frequencies are estimated from the smoothed speech spectrum by adding constraints on the formant frequency ranges. The four highest peaks are typically classified as the first four formants. However, the LPC algorithm estimate formant frequencies from the all pole model of the vocal tract transfer function. The approach relies on the source – filter m...
In this paper, two speech enhancement algorithms (SEAs) based on spectral subtraction (SS) princi... more In this paper, two speech enhancement algorithms (SEAs) based on spectral subtraction (SS) principle have been evaluated for bilateral cochlear implant (BCI) users. Specifically, dual-channel noise power spectral estimation algorithm using power spectral densities (PSD) and cross power spectral density (CPSD) of the observed signals was studied. The enhanced speech signals were obtained using either Dual Channel Non Linear Spectral Subtraction 'DC-NLSS' or Dual-Channel Multi-Band Spectral Subtraction 'DC-MBSS' algorithms. For performance evaluation, some objective speech assessment tests relying on Perceptual Evaluation of Speech Quality (PESQ) score and speech Itakura-Saito (IS) distortion measurement were performed to fix the optimal number of frequency band needed in DC-MBSS algorithm. In order to evaluate the speech intelligibility, subjective listening tests were assessed with 50 normal hearing listeners using a specific BCI simulator and with three deafened BCI patients. Experimental results, obtained using French Lafon database corrupted by an additive babble noise at different Signal-to-Noise Ratios (SNR), showed that DC-MBSS algorithm improves speech understanding better than DC-NLSS algorithm for single and multiple interfering noise sources.
2020 IEEE 4th International Conference on Image Processing, Applications and Systems (IPAS), 2020
Handwritten recognition has received considerable attention in the domain of pattern recognition,... more Handwritten recognition has received considerable attention in the domain of pattern recognition, image processing, over the last few decades. As a consequence of this research effort, several algorithms were developed using different techniques. Particularly, Deep Learning has shown a remarkable capability to handle handwritten recognition in very recent years. The well-known Deep learning techniques are the Convolutional Neuronal Networks (CNNs) and Recurrent Neuronal Networks (RNNs). This paper provides a survey of the most recent handwritten recognition systems. Thus, we present the most significant algorithms for handwritten character/word/text recognition by explaining the different approaches used in the recognition process and we compare them in terms of accuracy.
Knowledge Science, Engineering and Management, 2019
Speaker recognition systems proposed in the literature are conditioned by many factors such as th... more Speaker recognition systems proposed in the literature are conditioned by many factors such as the features used, the type of data to be processed, data duration and the approach to be used. This work presents a novel speaker recognition application using a new dimensional cepstral feature vector for building Gaussian Mixture Models (GMM) for speaker identification and verification systems. Recognition results obtained with the proposed system are compared to the state of the art systems requiring high dimensional feature vectors. We experimentally evaluate this work with speakers taken from TIMIT database. The new system gave substantial improvements even when trained with limited data, which explain the effectiveness of the new feature vectors with this approach which can be comparable to the use additional algorithms or approaches.
2020 International Wireless Communications and Mobile Computing (IWCMC), 2020
An important amount of speech is typically required for speaker identification system development... more An important amount of speech is typically required for speaker identification system development and evaluation. Nowadays, robust speaker identification systems when short utterances are used remains a key consideration for automatic speaker recognition, since a lot of real world applications are able to deal with only limited duration speech data. This paper presents a new approach based on a low complexity solution based on a new feature vectors to build Gaussian Mixture Models (GMM) for speaker identification systems especially when training and testing utterance lengths are reduced. We compared our proposed system to the state-of-the-art based system in Speaker identification. Experiments on TIMIT database were conducted to demonstrate that this new feature vector can outperform the standard GMM-based system and show that there is no need for extra-data to identify the speakers.
2021 International Wireless Communications and Mobile Computing (IWCMC), 2021
This paper presents a new text independent speaker recognition system based on new cepstral Coeff... more This paper presents a new text independent speaker recognition system based on new cepstral Coefficients and i-vector based on probabilistic linear discriminant analysis. This system is designed to be used as a biometric authentication system based on the speaker voice. The experiments were performed on speech data taken from a large number of users and consist of 630 speakers from TIMIT database for different training conditions. The efficiency of the proposed system is observed with the increase of Speaker identification Rate. Experimental results show that the system gives higher recognition performance for different numbers of speakers.
2017 International Conference on Advanced Technologies for Signal and Image Processing (ATSIP), 2017
Speaker identity, the sound of a person's voice, is one of the most important characteristics... more Speaker identity, the sound of a person's voice, is one of the most important characteristics in human communication. Voice conversion (VC) is an emergent problem in voice and speech processing that deals with the process of modifying a speaker's identity. More particularly, the speech signal spoken by the source speaker is modified to sound a sifit had been pronounced by another speaker, referred to as the target speaker. A variety of VC techniques has been proposed since the first appearance of the voice conversion problem. The choice among those techniques represents a compromise between the similarity of the converted voice to the target voice and the quality of the output speech signal, both rated by the used technique. In this paper, we review a comprehensive state-of-the-art of voice conversion techniques while pointing out their advantages and disadvantages. These techniques will be applied in significant and most versatile areas of speech technology; applications that are far beyond speech synthesis.
Acoustic analysis of speech signal enables automatic detection and classification of voice disord... more Acoustic analysis of speech signal enables automatic detection and classification of voice disorders along with its severity. This automatic assessment provides help to the clinician in initial diagnosis of pathological larynx in non-intrusive way. Voice pathologies damage the vocal cords and consequently alter the dynamics (fluctuation speed) of vocal cords. In this article, we have estimated glottal volume velocity waveform (GVVW) from the speech pressure waveforms of healthy and pathological subjects using quasi closed phase (QCP) glottal inverse filtering algorithm to capture altered dynamics of vocal cords. Closed-phase methods revealed notable stability in diverse voice qualities and sub-glottal pressures. The GVVW is the source of significant acoustical clues rooted in speech. The estimated GVVW is then parameterized by various time based, frequency based and Liljencrants-Fant (LF) model based glottal descriptors. Glottal descriptor's vectors have been passed on to stochastic gradient descent (SGD) classifier for voice disorder evaluation. The normal pitch utterance of sustained vowel /a/ quarried from German, English, Arabic and Spanish voice databases is used. Information gain (IG) feature scoring technique is employed to select optimal descriptors and to rank them. Several intra and cross-database experiments were performed to explore the usefulness of glottal descriptors for voice disorder detection, severity detection and classification. Student's t-tests were performed to validate the obtained results.
Recent advances in the speaker recognition (SR) field showed remarkably accurate and outperformin... more Recent advances in the speaker recognition (SR) field showed remarkably accurate and outperforming algorithms. However, their performances drastically degrade when the sparse amount of data is available. Nowadays, recognising a speaker identity when only a small amount of speech data is involved for testing and training remains a key consideration since many real world applications often have access to only speech data having a limited duration. In this study, the authors present a new improved approach, based on new information detected from the speech signal, to improve the task of automatic speaker identification. In doing so, they highlight how the detection of the speaker dialect can be explored to address the research problem related to short utterance SR. Results obtained with the new regional system are presented which provide a comparison between this system and the state-of-the-art systems for speaker identification task.
Uploads
Papers by Mondher Frikha