Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935584
Irfanudin Nor Anwar, K. Daud, A. Samat, Z. H. C. Soh, A. M. Omar, F. Ahmad
Power Quality Disturbances (PQD) has result in numerous failures and damage to electrical equipment. This paper utilized MATLAB Application to propose ways in detecting and classifying Voltage Sag, Swell and Transient. The proposal was divided into three parts which are detection, classification, and performance evaluation. The detection stage was done using Discrete Wavelet Transform in Wavelet Analyzer to obtain signal decomposition in different energy levels to be used in Energy Distribution Deviation (EDD) method. The classification stage was done in Classification Learner to check how good Multilayer Perceptron Neural Network able to trains, validates, and predicts as a classification model. The performance evaluation stage was done in Neural Net Fitting using Levenberg-Marquardt (LM) as training algorithm to see how well the model perform in term of Mean Square Error (MSE) and regression. This paper also discusses the effect of input ratio, activation function (Sigmoid, Tangent Hyperbolic, Rectified Linear Unit) and training algorithm (Levenberg-Marquardt, Bayesian Regularization, Scale Conjugate Gradient) towards accuracy in a Neural Network model. This study found that EDD was able to detect the difference in energy distribution of PQD properly. The Multilayer Perceptron model was observed to performed better and had higher accuracy when fed with more sample data, bigger layer size and activated using Tangent Hyperbolic (Tanh) activation function. Increasing layer size also resulted in slower prediction speed and longer training time. The model performance was evaluated with the lowest MSE and highest regression when Levenberg-Marquardt (LM) was implemented compared to Bayesian Regularization (BR) and Scale Conjugate Gradient (SCG).
{"title":"Implementation of Levenberg-Marquardt Based Multilayer Perceptron (MLP) for Detection and Classification of Power Quality Disturbances","authors":"Irfanudin Nor Anwar, K. Daud, A. Samat, Z. H. C. Soh, A. M. Omar, F. Ahmad","doi":"10.1109/ICCSCE54767.2022.9935584","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935584","url":null,"abstract":"Power Quality Disturbances (PQD) has result in numerous failures and damage to electrical equipment. This paper utilized MATLAB Application to propose ways in detecting and classifying Voltage Sag, Swell and Transient. The proposal was divided into three parts which are detection, classification, and performance evaluation. The detection stage was done using Discrete Wavelet Transform in Wavelet Analyzer to obtain signal decomposition in different energy levels to be used in Energy Distribution Deviation (EDD) method. The classification stage was done in Classification Learner to check how good Multilayer Perceptron Neural Network able to trains, validates, and predicts as a classification model. The performance evaluation stage was done in Neural Net Fitting using Levenberg-Marquardt (LM) as training algorithm to see how well the model perform in term of Mean Square Error (MSE) and regression. This paper also discusses the effect of input ratio, activation function (Sigmoid, Tangent Hyperbolic, Rectified Linear Unit) and training algorithm (Levenberg-Marquardt, Bayesian Regularization, Scale Conjugate Gradient) towards accuracy in a Neural Network model. This study found that EDD was able to detect the difference in energy distribution of PQD properly. The Multilayer Perceptron model was observed to performed better and had higher accuracy when fed with more sample data, bigger layer size and activated using Tangent Hyperbolic (Tanh) activation function. Increasing layer size also resulted in slower prediction speed and longer training time. The model performance was evaluated with the lowest MSE and highest regression when Levenberg-Marquardt (LM) was implemented compared to Bayesian Regularization (BR) and Scale Conjugate Gradient (SCG).","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132510504","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935654
Aiman Hamizan Tuan Rusli, Belinda Chong Chiew Meng, N. S. Damanhuri, N. A. Othman, Mohamad Haizan Othman, Wan Fatimah Azzahra Wan Zaidi
Agricultural production is one of the main sources of income in most countries. Enormous losses will be incurred if agricultural product is disturbed by plant disease. The key to reduce losses in agricultural product output and quantity is early detection of plant diseases. A diseased plant usually reflecting its disease by showing symptoms on its leaves. A potato leaf disease classification technique by using image processing and artificial neural network method is proposed in this study. The method can be used to determine the potato leaf is either healthy or diseased. With the aid of this technique, farmers can save time and cost in their farming activities. The main goal of this study is to detect potato plant (Solanum tuberosum L.) disease using image processing techniques. The K-Means clustering algorithm is used to segment the disease in potato leaf image. The segmented features of potato leaf disease are then extracted by using Gray Level Co-occurrence Matrix (GLCM) and these features are then fed into ANN for classification. With the proposed system, classification accuracy obtained is 94%.
{"title":"Potato Leaf Disease Classification using Image Processing and Artificial Neural Network","authors":"Aiman Hamizan Tuan Rusli, Belinda Chong Chiew Meng, N. S. Damanhuri, N. A. Othman, Mohamad Haizan Othman, Wan Fatimah Azzahra Wan Zaidi","doi":"10.1109/ICCSCE54767.2022.9935654","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935654","url":null,"abstract":"Agricultural production is one of the main sources of income in most countries. Enormous losses will be incurred if agricultural product is disturbed by plant disease. The key to reduce losses in agricultural product output and quantity is early detection of plant diseases. A diseased plant usually reflecting its disease by showing symptoms on its leaves. A potato leaf disease classification technique by using image processing and artificial neural network method is proposed in this study. The method can be used to determine the potato leaf is either healthy or diseased. With the aid of this technique, farmers can save time and cost in their farming activities. The main goal of this study is to detect potato plant (Solanum tuberosum L.) disease using image processing techniques. The K-Means clustering algorithm is used to segment the disease in potato leaf image. The segmented features of potato leaf disease are then extracted by using Gray Level Co-occurrence Matrix (GLCM) and these features are then fed into ANN for classification. With the proposed system, classification accuracy obtained is 94%.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133964645","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935637
Muhammad Aliff Rosly, H. Yussof, Svamimi Shamsuddin, N. I. Zahari, Ahmad Zamir Che Daud
Eye-tracking is regarded as a valuable instrument for evaluating intervention programmes, especially those in the social or communication categories. It includes the robot-mediated intervention in which a robot is utilised to converse with children during therapy. Nevertheless, recent robot-mediated interventions continue to measure eye contact manually using video recordings for evaluation purposes. Using an additional measuring device other than the robot itself is inefficient without exploring its advanced robotics capabilities. Therefore, this research suggests measuring eye contact using an NAO robot vision and compares it to the conventional recorded video analysis. During a therapy session, the NAO robot's cameras automatically measure and compute eye contact data. The NAOqi PeoplePerception ALGazeAnalysis API analyses the detected individual's gaze direction. The ‘look’ and ‘not look’ events are alternately raised till the end of the module time, with each eye contact duration added to the total sum for calculation. The code has been improved to account for unnecessary detection during momentary eye contact aversion or glance for a more accurate assessment. Then, an experiment is undertaken to compare the measurement to the traditional recorded video approach at each range. The ON difference data were plotted on a Bland-Altman graph to determine the degree of agreement between the two approaches. Even their 95 per cent confidence intervals fall well inside the maximum variance allowed. This indicates that both methods demonstrate excellent agreement, and there is no noticeable difference between them. Consequently, it may be argued that the NAO robot can replace the traditional recorded methodology or that the two methods are interchangeable.
眼动追踪被认为是评估干预方案的一种有价值的工具，特别是在社会或交流领域。它包括机器人介导的干预，在治疗期间，机器人被用来与儿童交谈。然而，最近的机器人干预仍然是为了评估目的而使用视频记录手动测量目光接触。使用机器人本身以外的额外测量设备是低效的，而不探索其先进的机器人功能。因此，本研究建议使用NAO机器人视觉来测量目光接触，并将其与传统的录制视频分析进行比较。在治疗过程中，NAO机器人的摄像头会自动测量和计算眼神交流数据。NAOqi PeoplePerception algeanalysis API分析被检测个体的凝视方向。' look '和' not look '事件交替引发，直到模块时间结束，每次目光接触持续时间都被添加到计算的总和中。该代码已得到改进，以解释在瞬间眼神接触厌恶或一瞥期间不必要的检测，以进行更准确的评估。然后，在每个范围内进行实验，将测量结果与传统的录制视频方法进行比较。ON差异数据绘制在Bland-Altman图上，以确定两种方法之间的一致程度。即使它们95%的置信区间也完全落在允许的最大方差之内。这表明两种方法具有很好的一致性，两者之间没有明显的差异。因此，可以认为NAO机器人可以取代传统的记录方法，或者这两种方法是可互换的。
{"title":"Eye Contact Measurement using NAO Robot Vision for Autism Intervention","authors":"Muhammad Aliff Rosly, H. Yussof, Svamimi Shamsuddin, N. I. Zahari, Ahmad Zamir Che Daud","doi":"10.1109/ICCSCE54767.2022.9935637","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935637","url":null,"abstract":"Eye-tracking is regarded as a valuable instrument for evaluating intervention programmes, especially those in the social or communication categories. It includes the robot-mediated intervention in which a robot is utilised to converse with children during therapy. Nevertheless, recent robot-mediated interventions continue to measure eye contact manually using video recordings for evaluation purposes. Using an additional measuring device other than the robot itself is inefficient without exploring its advanced robotics capabilities. Therefore, this research suggests measuring eye contact using an NAO robot vision and compares it to the conventional recorded video analysis. During a therapy session, the NAO robot's cameras automatically measure and compute eye contact data. The NAOqi PeoplePerception ALGazeAnalysis API analyses the detected individual's gaze direction. The ‘look’ and ‘not look’ events are alternately raised till the end of the module time, with each eye contact duration added to the total sum for calculation. The code has been improved to account for unnecessary detection during momentary eye contact aversion or glance for a more accurate assessment. Then, an experiment is undertaken to compare the measurement to the traditional recorded video approach at each range. The ON difference data were plotted on a Bland-Altman graph to determine the degree of agreement between the two approaches. Even their 95 per cent confidence intervals fall well inside the maximum variance allowed. This indicates that both methods demonstrate excellent agreement, and there is no noticeable difference between them. Consequently, it may be argued that the NAO robot can replace the traditional recorded methodology or that the two methods are interchangeable.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130052197","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935651
Nur Nabilah Bahrum, S. Setumin, Edi Afzan Saidon, N. A. Othman, M. F. Abdullah
In law enforcement, face sketch recognition has been used to identify the criminal suspect. Usually, when there is no other evidence, a forensic artist will draw the face of the suspect based on the eyewitness description. Then, the forensic sketch will be matched with the mugshot images from the database in order to recognize and identify the potential suspect. However, the matching performance of the forensic sketches could be affected by various factors, and one of the major factors is the occlusion that exists in the sketch itself. This is because most of the suspects usually wear something that could help in hiding their identities, like a face mask, glasses, hoodie, or cap, when they are committing a crime. Since the mugshot images do not include the occlusion, it will make it harder to recognize the suspect in the matching process, even if the sketch and photo are from the same person. This is due to the larger Euclidean distance between the extracted features from these two images, particularly in the occlusion regions. Therefore, this study proposed a method that matches only the pre-selected regions that exclude occlusion in both images. This region of interest is pre-selected on the forensic face sketch before the same region is applied to all mugshot images. In this study, the forensic sketch with their corresponding photo was obtained from the PRIP-HDC dataset, and the Histogram of Gradient (HOG) was used for feature extraction. Based on the result obtained, this study's performance shows some improvement in recognizing the forensic sketches compared to the existing technique.
{"title":"Forensic Face Sketch Recognition based on Pre-Selected Facial Regions","authors":"Nur Nabilah Bahrum, S. Setumin, Edi Afzan Saidon, N. A. Othman, M. F. Abdullah","doi":"10.1109/ICCSCE54767.2022.9935651","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935651","url":null,"abstract":"In law enforcement, face sketch recognition has been used to identify the criminal suspect. Usually, when there is no other evidence, a forensic artist will draw the face of the suspect based on the eyewitness description. Then, the forensic sketch will be matched with the mugshot images from the database in order to recognize and identify the potential suspect. However, the matching performance of the forensic sketches could be affected by various factors, and one of the major factors is the occlusion that exists in the sketch itself. This is because most of the suspects usually wear something that could help in hiding their identities, like a face mask, glasses, hoodie, or cap, when they are committing a crime. Since the mugshot images do not include the occlusion, it will make it harder to recognize the suspect in the matching process, even if the sketch and photo are from the same person. This is due to the larger Euclidean distance between the extracted features from these two images, particularly in the occlusion regions. Therefore, this study proposed a method that matches only the pre-selected regions that exclude occlusion in both images. This region of interest is pre-selected on the forensic face sketch before the same region is applied to all mugshot images. In this study, the forensic sketch with their corresponding photo was obtained from the PRIP-HDC dataset, and the Histogram of Gradient (HOG) was used for feature extraction. Based on the result obtained, this study's performance shows some improvement in recognizing the forensic sketches compared to the existing technique.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130219438","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935643
Allen Lois Lanuza, Roxanne De Leon, C. R. Lucas
Peak latency measurement of the patient's Auditory Brainstem Response (ABR) essential wave components (Waves I-V) is the usual method in hearing screening to determine the likelihood of hearing impairment. To visualize the peaks of Waves I-V, averaging about 2000 ABR sweeps is necessary for reducing the background noise caused by power line interference and myogenic activity; however, this method is time-consuming and inconvenient for patients and healthcare workers. The study aims to use signal denoising methods to denoise ABRs averaged with fewer sweeps without affecting their functionality. Two deterministic signal denoising approaches, Empirical Mode Decomposition (EMD) and Discrete Wavelet Transform (DWT), were evaluated and compared to determine which could produce functional denoised ABRs using fewer sweeps. For the 1 kHz stimulus frequency, DWT produced functional ABRs with fewer sweeps than EMD for stimulus intensities of 75, 65, 55 and 50 dB peSPL. For the 4 kHz stimulus frequency, only the DWT method could produce functional ABRs with fewer sweeps. DWT method performs better than EMD in producing clinically relevant denoised ABR for most stimulus descriptions. The findings can help audiologists use the DWT denoising approach when averaging noisy ABRs with fewer sweeps to address the problems caused by the time-consuming conventional averaging method.
听觉脑干反应(ABR)基本波分量(波I-V)的峰值潜伏期测量是听力筛查中确定听力障碍可能性的常用方法。为了可视化波I-V的峰值，平均约2000 ABR扫描是必要的，以减少由电力线干扰和肌生成活动引起的背景噪声;然而，这种方法耗时长，对患者和医护人员不方便。本研究旨在利用信号去噪方法，在不影响abr功能的情况下，对扫描次数较少的平均abr进行去噪。对两种确定性信号去噪方法——经验模态分解(EMD)和离散小波变换(DWT)进行了评估和比较，以确定哪种方法可以使用更少的扫描产生功能去噪的abr。对于1 kHz的刺激频率，在75、65、55和50 dB peSPL的刺激强度下，DWT产生的功能性abr比EMD产生的扫描次数少。对于4 kHz的刺激频率，只有DWT方法可以产生较少扫描次数的功能性abr。对于大多数刺激描述，DWT方法在产生与临床相关的去噪ABR方面优于EMD方法。该研究结果可以帮助听力学家使用DWT去噪方法，以较少的扫描次数平均有噪声的abr，以解决耗时的传统平均方法所带来的问题。
{"title":"Comparative Analysis of Empirical Mode Decomposition and Discrete Wavelet Transform as Denoising Methods for Auditory Brainstem Response","authors":"Allen Lois Lanuza, Roxanne De Leon, C. R. Lucas","doi":"10.1109/ICCSCE54767.2022.9935643","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935643","url":null,"abstract":"Peak latency measurement of the patient's Auditory Brainstem Response (ABR) essential wave components (Waves I-V) is the usual method in hearing screening to determine the likelihood of hearing impairment. To visualize the peaks of Waves I-V, averaging about 2000 ABR sweeps is necessary for reducing the background noise caused by power line interference and myogenic activity; however, this method is time-consuming and inconvenient for patients and healthcare workers. The study aims to use signal denoising methods to denoise ABRs averaged with fewer sweeps without affecting their functionality. Two deterministic signal denoising approaches, Empirical Mode Decomposition (EMD) and Discrete Wavelet Transform (DWT), were evaluated and compared to determine which could produce functional denoised ABRs using fewer sweeps. For the 1 kHz stimulus frequency, DWT produced functional ABRs with fewer sweeps than EMD for stimulus intensities of 75, 65, 55 and 50 dB peSPL. For the 4 kHz stimulus frequency, only the DWT method could produce functional ABRs with fewer sweeps. DWT method performs better than EMD in producing clinically relevant denoised ABR for most stimulus descriptions. The findings can help audiologists use the DWT denoising approach when averaging noisy ABRs with fewer sweeps to address the problems caused by the time-consuming conventional averaging method.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117090457","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935642
A. A. Jamil, J. Kadir, Johanis Mohd Jamil, F.R. Hashim, S. Shaharuddin, Nazrul Fariq Makmor
The development of artificial neural networks (ANNs) was founded on computer alterations of human biology (the concept of neurons). The practicality of applying ANNs to various problems has been the subject of numerous studies, particularly in the field of biomedical engineering. Medical and educational decision-making regularly use applications to ANNs. Using a range of reference data, the ANNs used in the current study were trained to recognise cardiac abnormalities. Typically referred to as reference parameters, electrocardiogram (ECG) signal amplitude and duration are employed as input parameters for cardiac issues. An ECG complex consists of a P peak, QRS wave, and T peak. The amplitude and length of each P peak, QRS wave, and T peak are measured, resulting in a total of six input parameters for the artificial neural network. The artificial neural network (ANN) structure in this study is a multilayer perceptron (MLP), and the training techniques are Bayesian Regularization (BayR), Lavenberg Marquardt (LevM), and Backpropagation (BackP). The influence of the Tansig activation function on the MLP structure. The MLP network that achieved the highest accuracy (94.44%) utilising the BayR training method and Logsig activation function surpassed all others.
{"title":"Multilayer Perceptron Optimization of ECG Peaks for Cardiac Abnormality Detection","authors":"A. A. Jamil, J. Kadir, Johanis Mohd Jamil, F.R. Hashim, S. Shaharuddin, Nazrul Fariq Makmor","doi":"10.1109/ICCSCE54767.2022.9935642","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935642","url":null,"abstract":"The development of artificial neural networks (ANNs) was founded on computer alterations of human biology (the concept of neurons). The practicality of applying ANNs to various problems has been the subject of numerous studies, particularly in the field of biomedical engineering. Medical and educational decision-making regularly use applications to ANNs. Using a range of reference data, the ANNs used in the current study were trained to recognise cardiac abnormalities. Typically referred to as reference parameters, electrocardiogram (ECG) signal amplitude and duration are employed as input parameters for cardiac issues. An ECG complex consists of a P peak, QRS wave, and T peak. The amplitude and length of each P peak, QRS wave, and T peak are measured, resulting in a total of six input parameters for the artificial neural network. The artificial neural network (ANN) structure in this study is a multilayer perceptron (MLP), and the training techniques are Bayesian Regularization (BayR), Lavenberg Marquardt (LevM), and Backpropagation (BackP). The influence of the Tansig activation function on the MLP structure. The MLP network that achieved the highest accuracy (94.44%) utilising the BayR training method and Logsig activation function surpassed all others.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114208299","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935581
Jomari B. Ganhinhin, Maria Donnabelle B. Varona, C. R. Lucas, Angelina A. Aquino
Existing Tagalog Text-to-speech (TTS) systems still have room for improvement, and although recent attempts at creating local TTS systems for Philippine spoken languages were able to generate synthesized speech, they still possess relatively low Mean Opinion Scores (MOS), ranging from 1.5 to 3.9 (out of 5), when it comes to naturalness and intelligibility. Improving speech prosody, the main factor for a speech's naturalness or individuality, has been made possible through voice conversion (VC). This project aims to implement a VC system for Tagalog synthesized speech, specifically using Cycle Generative Adversarial Networks (Cycle-GAN), a state-of-the-art neural network architecture used in non-parallel VC. Inter-gender and intra-gender VC were made for two types of inputs: Google's own Tagalog TTS and a locally sourced TTS system built from Mary TTS. Results show that Google TTS and its VC models perform better overall than Mary TTS and its VC models. Mel Cepstral Distortions (MCD) and F0: Root Mean Square Errors (F0:RMSE) vary across all models, reaching an MCD as low as 6.52 dB for Google TTS' intra-gender VC and an F0:RMSE as low as 16.92 Hz from Google TTS' inter-gender VC. Meanwhile, undergoing VC also caused a degradation in perceived speech quality as seen in a decrease in MOS across all VC models. Inter-gender VC for both TTS inputs were subjectively more preferred over intra-gender VC, reaching MOS values of 3.76 and 2.32 for Google TTS and Mary TTS inputs, respectively. Furthermore, it was also shown that male respondents were likely to rate higher opinion scores for intra-gender VC than female respondents, likely due to differences in hearing sensitivities.
{"title":"Voice Conversion of Tagalog Synthesized Speech Using Cycle-Generative Adversarial Networks (Cycle-GAN)","authors":"Jomari B. Ganhinhin, Maria Donnabelle B. Varona, C. R. Lucas, Angelina A. Aquino","doi":"10.1109/ICCSCE54767.2022.9935581","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935581","url":null,"abstract":"Existing Tagalog Text-to-speech (TTS) systems still have room for improvement, and although recent attempts at creating local TTS systems for Philippine spoken languages were able to generate synthesized speech, they still possess relatively low Mean Opinion Scores (MOS), ranging from 1.5 to 3.9 (out of 5), when it comes to naturalness and intelligibility. Improving speech prosody, the main factor for a speech's naturalness or individuality, has been made possible through voice conversion (VC). This project aims to implement a VC system for Tagalog synthesized speech, specifically using Cycle Generative Adversarial Networks (Cycle-GAN), a state-of-the-art neural network architecture used in non-parallel VC. Inter-gender and intra-gender VC were made for two types of inputs: Google's own Tagalog TTS and a locally sourced TTS system built from Mary TTS. Results show that Google TTS and its VC models perform better overall than Mary TTS and its VC models. Mel Cepstral Distortions (MCD) and F0: Root Mean Square Errors (F0:RMSE) vary across all models, reaching an MCD as low as 6.52 dB for Google TTS' intra-gender VC and an F0:RMSE as low as 16.92 Hz from Google TTS' inter-gender VC. Meanwhile, undergoing VC also caused a degradation in perceived speech quality as seen in a decrease in MOS across all VC models. Inter-gender VC for both TTS inputs were subjectively more preferred over intra-gender VC, reaching MOS values of 3.76 and 2.32 for Google TTS and Mary TTS inputs, respectively. Furthermore, it was also shown that male respondents were likely to rate higher opinion scores for intra-gender VC than female respondents, likely due to differences in hearing sensitivities.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116966207","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935657
Z. Muhammad, Nur Aqilah Jak Jailani, N. A. M. Leh, S. A. Hamid
Water is extremely important in both the environmental and social realms. The consumption of clean water guarantees a quality of life as it provides essential minerals and nutrients to the body. Water pollution posing a threat to human health, ecosystems, plant, and animal life. Today, Malaysia is showing an increasing rate of water pollution as there are currently undergoing tremendous urbanization and population expansion. The Water Quality Index (WQI) must monitor frequently to ensure the level of water cleanliness and safeness. However, monitoring work was conduct manually are time consuming, requires a lot of manpower and high expertise in determining the level of water cleanliness. Due to those issues, the intention of this study is to develop an automatic method in water quality classification for drinking purpose whether it is potable or non-potable using Support Vector Machine (SVM) which is more accurate, fast, and easy. This project used up to 59 samples of data from various location to prepare the SVM with two different kernels. By using MATLAB version R2021A, the implementation of this project was performed. Based on the result obtained, it is discovered that SVM model with RBF kernel has the better performance with high percentage of accuracy, precision, sensitivity, and specificity compared to SVM model with Polynomial kernel. All two types of kernels were accepted to be used in SVM model water quality classifier as their performance's criteria which are accuracy, specificity, sensitivity, and precision were greater than 80%. The findings of the study were benefits to the other or future work, particularly in the water quality classification system.
{"title":"Classification of Drinking Water Quality using Support Vector Machine (SVM) Algorithm","authors":"Z. Muhammad, Nur Aqilah Jak Jailani, N. A. M. Leh, S. A. Hamid","doi":"10.1109/ICCSCE54767.2022.9935657","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935657","url":null,"abstract":"Water is extremely important in both the environmental and social realms. The consumption of clean water guarantees a quality of life as it provides essential minerals and nutrients to the body. Water pollution posing a threat to human health, ecosystems, plant, and animal life. Today, Malaysia is showing an increasing rate of water pollution as there are currently undergoing tremendous urbanization and population expansion. The Water Quality Index (WQI) must monitor frequently to ensure the level of water cleanliness and safeness. However, monitoring work was conduct manually are time consuming, requires a lot of manpower and high expertise in determining the level of water cleanliness. Due to those issues, the intention of this study is to develop an automatic method in water quality classification for drinking purpose whether it is potable or non-potable using Support Vector Machine (SVM) which is more accurate, fast, and easy. This project used up to 59 samples of data from various location to prepare the SVM with two different kernels. By using MATLAB version R2021A, the implementation of this project was performed. Based on the result obtained, it is discovered that SVM model with RBF kernel has the better performance with high percentage of accuracy, precision, sensitivity, and specificity compared to SVM model with Polynomial kernel. All two types of kernels were accepted to be used in SVM model water quality classifier as their performance's criteria which are accuracy, specificity, sensitivity, and precision were greater than 80%. The findings of the study were benefits to the other or future work, particularly in the water quality classification system.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122887168","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935659
Safyzan Salim, M. M. A. Jamil, R. Ambar, R. Roslan, M. G. Kamardan
A major challenge when developing Machine Learning (ML) sign language recognition using wearable is how to efficiently translate the gestures based on the acquired sensors data. Conventional method utilizes data fusion based on the obtained sensors' information by producing mapping/lookup table for creating classification model of gestures corresponding sensor value. Although this method is effective, it increases programming complexity. Therefore, emerging technology that can improve the simplicity and provide accuracy of gestures' data processing is needed. This work experiments the artificial intelligence approach of the development of American Sign Language (ASL) detection using MediaPipe, a ready-to-use cross-platform machine learning framework for computer vision works and Google Teachable Machine a free web tool of machine learning model creation.
{"title":"Sign Language Digit Detection with MediaPipe and Machine Learning Algorithm","authors":"Safyzan Salim, M. M. A. Jamil, R. Ambar, R. Roslan, M. G. Kamardan","doi":"10.1109/ICCSCE54767.2022.9935659","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935659","url":null,"abstract":"A major challenge when developing Machine Learning (ML) sign language recognition using wearable is how to efficiently translate the gestures based on the acquired sensors data. Conventional method utilizes data fusion based on the obtained sensors' information by producing mapping/lookup table for creating classification model of gestures corresponding sensor value. Although this method is effective, it increases programming complexity. Therefore, emerging technology that can improve the simplicity and provide accuracy of gestures' data processing is needed. This work experiments the artificial intelligence approach of the development of American Sign Language (ASL) detection using MediaPipe, a ready-to-use cross-platform machine learning framework for computer vision works and Google Teachable Machine a free web tool of machine learning model creation.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121061847","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-10-21DOI: 10.1109/ICCSCE54767.2022.9935627
Muhammad Adib Faidhi Daud, A. P. Ismail, N. Tahir, K. Daud, Nazirah Mohamat Kasim, Fadzil Ahmad Mohamad
Drowsy driving is one of the most common causes of road accidents. Human usually become drowsy when tired and it is dangerous especially during driving on the road. Drowsiness can induce microsleep which can cause a significant decline in driving performance and thus would increase the chance of accidents. Hence, this real time drowsy driver detection is developed that to help minimize the chance of road accidents occurrence when the driver become drowsy. In this proposed method, the drowsy driver can be detected and alerted without using any intrusive instruments that could distract the driver. This drowsy detection is done using real time input image of the driver using a camera and image processing using Python. Next, drowsiness sign can be detected from the facial expression of the driver through the percentage of eyes opened and the frequent yawning. From the facial expression, the calculation of the eye closure known as eye aspect ratio (EAR) and the wideness of mouth opening known as mouth aspect ratio (MAR) can be made. Finally, using the value obtained, the system can determine whether the driver is alert or drowsy.
{"title":"Real Time Drowsy Driver Detection Using Image Processing on Python","authors":"Muhammad Adib Faidhi Daud, A. P. Ismail, N. Tahir, K. Daud, Nazirah Mohamat Kasim, Fadzil Ahmad Mohamad","doi":"10.1109/ICCSCE54767.2022.9935627","DOIUrl":"https://doi.org/10.1109/ICCSCE54767.2022.9935627","url":null,"abstract":"Drowsy driving is one of the most common causes of road accidents. Human usually become drowsy when tired and it is dangerous especially during driving on the road. Drowsiness can induce microsleep which can cause a significant decline in driving performance and thus would increase the chance of accidents. Hence, this real time drowsy driver detection is developed that to help minimize the chance of road accidents occurrence when the driver become drowsy. In this proposed method, the drowsy driver can be detected and alerted without using any intrusive instruments that could distract the driver. This drowsy detection is done using real time input image of the driver using a camera and image processing using Python. Next, drowsiness sign can be detected from the facial expression of the driver through the percentage of eyes opened and the frequent yawning. From the facial expression, the calculation of the eye closure known as eye aspect ratio (EAR) and the wideness of mouth opening known as mouth aspect ratio (MAR) can be made. Finally, using the value obtained, the system can determine whether the driver is alert or drowsy.","PeriodicalId":346014,"journal":{"name":"2022 IEEE 12th International Conference on Control System, Computing and Engineering (ICCSCE)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122438506","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：