Dittaya Wanvarie
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Dittaya Wanvarie has a total of 42 co-authors. Here are the top co-authors by number of shared publications:
- Yiping Jin
Department of Mathematics and Computer Science Faculty of Science Chulalongkorn University
6 shared publications Phu T. V. Le
Knorex, 140 Robinson Road, #14-16 Crown @ Robinson, Singapore 068907
4 shared publications- Ploy Naruemon Pratanwanich
Chulalongkorn University
4 shared publications - Proadpran Punyabukkana
Chulalongkorn University
4 shared publications - Chaipat Chunharas
Cognitive Clinical and Computational Neuroscience Center of Excellence, Department of Medicine, King Chulalongkorn Memorial Hospital, Chulalongkorn University
2 shared publications - Masashi Sugiyama
RIKEN Center for Advanced Intelligence Project
2 shared publications
Latest Publications
- Open Access
- 5 December 2025
- Published by Springer Nature in Research Square
Enhancing Parkinson’s Diagnosis with PhonatoryVoice Analysis: Optimizing Machine LearningFeature Selection for Support Vector Machines
Parkinson’s disease (PD) is a common neurological disorder that impairs mobility and significantly impacts patients’ quality of life. Early diagnosis of PD is crucial for timely intervention and management. This study explores the potential of computer-aided analysis, specifically machine learning, for early PD diagnosis, focusing on phonatory voice analysis using dysphonia measures from sustained vowels. To enhance PD diagnosis performances, this study investigates the effectiveness of two feature selection methods, L1-norm support vector machine (L1-norm SVM) and Recursive Feature Elimination (RFE), within the framework of Support Vector Machine (SVM) and data from UCI Machine Learning Repository and Chulalongkorn Centre of Excellence for Parkinson’s Disease \& Related Disorders, a leading medical center for treating PD patients in Thailand, were utilized. Furthermore, this research uses PD voice samples to analyze dysphonia features for diagnosing PD. The results suggest that both L1-norm SVM and RFE enhance the accuracy of PD diagnosis, with L1-norm SVM showing higher performance, achieving an accuracy of 95.97%, a positive recall score of 97.16%, and an F1-macro score of 95.69%. The study also highlights the importance of key dysphonia measures, including RPDE, DFA, and HNR, in differentiating PD voices from non-PD voices. Further research is needed to explore additional feature selection methods, audio features, and datasets, which would help uncover more characteristics in early PD voice analysis and further enhance the performance of PD diagnosis.
- 5 December 2025
- Published by Springer Nature in Research Square
Open AccessEnhancing Parkinson’s Diagnosis with PhonatoryVoice Analysis: Optimizing Machine LearningFeature Selection for Support Vector Machines
Parkinson’s disease (PD) is a common neurological disorder that impairs mobility and significantly impacts patients’ quality of life. Early diagnosis of PD is crucial for timely intervention and management. This study explores the potential of computer-aided analysis, specifically machine learning, for early PD diagnosis, focusing on phonatory voice analysis using dysphonia measures from sustained vowels. To enhance PD diagnosis performances, this study investigates the effectiveness of two feature selection methods, L1-norm support vector machine (L1-norm SVM) and Recursive Feature Elimination (RFE), within the framework of Support Vector Machine (SVM) and data from UCI Machine Learning Repository and Chulalongkorn Centre of Excellence for Parkinson’s Disease \& Related Disorders, a leading medical center for treating PD patients in Thailand, were utilized. Furthermore, this research uses PD voice samples to analyze dysphonia features for diagnosing PD. The results suggest that both L1-norm SVM and RFE enhance the accuracy of PD diagnosis, with L1-norm SVM showing higher performance, achieving an accuracy of 95.97%, a positive recall score of 97.16%, and an F1-macro score of 95.69%. The study also highlights the importance of key dysphonia measures, including RPDE, DFA, and HNR, in differentiating PD voices from non-PD voices. Further research is needed to explore additional feature selection methods, audio features, and datasets, which would help uncover more characteristics in early PD voice analysis and further enhance the performance of PD diagnosis.
- Open Access
- 1 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 49306-49328
Efficient Text Bounding Box Identification Using Mask R-CNN: Case of Thai Documents
Text detection is a fundamental task in computer vision, particularly for Optical Character Recognition (OCR) applications. This study focuses on text detection within an OCR application, encompassing text detection, text recognition, and information extraction, with a specific emphasis on text detection. Character-Region Awareness for Text Detection (CRAFT), Pyramid Mask Text Detector (PMTD), and Scene Text Detection with Supervised Pyramid Context Network (SPCNET) have demonstrated promising results in bounding-box detection. However, it faces challenges related to postprocessing and multiline text detection. A post-processing problem arises because of the need to reconfigure the model when new documents are introduced, which leads to inefficiencies and complexities. In addition, especially for CRAFT tends to merge bounding boxes from consecutive lines by introducing multiline errors. To address these challenges, this study proposes an adapted approach based on Mask R-CNN, an instance segmentation model that treats each text element as an individual object. By adopting the Mask R-CNN approach, post-processing issues were successfully eliminated. Moreover, the multiline problem is effectively resolved. Comparative experiments demonstrate that the proposed model achieves comparable results to these models, while surpassing it in terms of accuracy and versatility. The proposed model is extensively evaluated on various document types, including bankbooks, Thai ID cards (both front and back sides), invoices, car registrations, mobile banking slips, passports, Indonesian ID cards, driver licenses, and receipts. The results indicated the model’s high performance and potential for real-world applications. The elimination of post-processing and multiline problems ensures the adaptability of the model to a wide range of document structures, and reduces both time inference and resource utilization.
- 1 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 49306-49328
Open AccessEfficient Text Bounding Box Identification Using Mask R-CNN: Case of Thai Documents
Text detection is a fundamental task in computer vision, particularly for Optical Character Recognition (OCR) applications. This study focuses on text detection within an OCR application, encompassing text detection, text recognition, and information extraction, with a specific emphasis on text detection. Character-Region Awareness for Text Detection (CRAFT), Pyramid Mask Text Detector (PMTD), and Scene Text Detection with Supervised Pyramid Context Network (SPCNET) have demonstrated promising results in bounding-box detection. However, it faces challenges related to postprocessing and multiline text detection. A post-processing problem arises because of the need to reconfigure the model when new documents are introduced, which leads to inefficiencies and complexities. In addition, especially for CRAFT tends to merge bounding boxes from consecutive lines by introducing multiline errors. To address these challenges, this study proposes an adapted approach based on Mask R-CNN, an instance segmentation model that treats each text element as an individual object. By adopting the Mask R-CNN approach, post-processing issues were successfully eliminated. Moreover, the multiline problem is effectively resolved. Comparative experiments demonstrate that the proposed model achieves comparable results to these models, while surpassing it in terms of accuracy and versatility. The proposed model is extensively evaluated on various document types, including bankbooks, Thai ID cards (both front and back sides), invoices, car registrations, mobile banking slips, passports, Indonesian ID cards, driver licenses, and receipts. The results indicated the model’s high performance and potential for real-world applications. The elimination of post-processing and multiline problems ensures the adaptability of the model to a wide range of document structures, and reduces both time inference and resource utilization.
- Open Access
- 4 November 2022
- Published by Elsevier in The Lancet Regional Health - Southeast Asia
- Vol. 8, 100106
Breakthrough infections, hospital admissions, and mortality after major COVID-19 vaccination profiles: a prospective cohort study
- 4 November 2022
- Published by Elsevier in The Lancet Regional Health - Southeast Asia
- Vol. 8, 100106
Open AccessBreakthrough infections, hospital admissions, and mortality after major COVID-19 vaccination profiles: a prospective cohort study
- Open Access
Selecting the Most Important Features for Predicting Mild Cognitive Impairment from Thai Verbal Fluency Assessments
Mild cognitive impairment (MCI) is an early stage of cognitive decline or memory loss, commonly found among the elderly. A phonemic verbal fluency (PVF) task is a standard cognitive test that participants are asked to produce words starting with given letters, such as “F” in English and “ก” /k/ in Thai. With state-of-the-art machine learning techniques, features extracted from the PVF data have been widely used to detect MCI. The PVF features, including acoustic features, semantic features, and word grouping, have been studied in many languages but not Thai. However, applying the same PVF feature extraction methods used in English to Thai yields unpleasant results due to different language characteristics. This study performs analytical feature extraction on Thai PVF data to classify MCI patients. In particular, we propose novel approaches to extract features based on phonemic clustering (ability to cluster words by phonemes) and switching (ability to shift between clusters) for the Thai PVF data. The comparison results of the three classifiers revealed that the support vector machine performed the best with an area under the receiver operating characteristic curve (AUC) of 0.733 (N = 100). Furthermore, our implemented guidelines extracted efficient features, which support the machine learning models regarding MCI detection on Thai PVF data.
Open AccessSelecting the Most Important Features for Predicting Mild Cognitive Impairment from Thai Verbal Fluency Assessments
Mild cognitive impairment (MCI) is an early stage of cognitive decline or memory loss, commonly found among the elderly. A phonemic verbal fluency (PVF) task is a standard cognitive test that participants are asked to produce words starting with given letters, such as “F” in English and “ก” /k/ in Thai. With state-of-the-art machine learning techniques, features extracted from the PVF data have been widely used to detect MCI. The PVF features, including acoustic features, semantic features, and word grouping, have been studied in many languages but not Thai. However, applying the same PVF feature extraction methods used in English to Thai yields unpleasant results due to different language characteristics. This study performs analytical feature extraction on Thai PVF data to classify MCI patients. In particular, we propose novel approaches to extract features based on phonemic clustering (ability to cluster words by phonemes) and switching (ability to shift between clusters) for the Thai PVF data. The comparison results of the three classifiers revealed that the support vector machine performed the best with an area under the receiver operating characteristic curve (AUC) of 0.733 (N = 100). Furthermore, our implemented guidelines extracted efficient features, which support the machine learning models regarding MCI detection on Thai PVF data.
- Open Access
Using Automatic Speech Recognition to Assess Thai Speech Language Fluency in the Montreal Cognitive Assessment (MoCA)
The Montreal cognitive assessment (MoCA), a widely accepted screening tool for identifying patients with mild cognitive impairment (MCI), includes a language fluency test of verbal functioning; its scores are based on the number of unique correct words produced by the test taker. However, it is possible that unique words may be counted differently for various languages. This study focuses on Thai as a language that differs from English in terms of word combinations. We applied various automatic speech recognition (ASR) techniques to develop an assisted scoring system for the MoCA language fluency test with Thai language support. This was a challenge because Thai is a low-resource language for which domain-specific data are not publicly available, especially speech data from patients with MCIs. Furthermore, the great variety of pronunciation, intonation, tone, and accent of the patients, all of which might differ from healthy controls, bring more complexity to the model. We propose a hybrid time delay neural network hidden Markov model (TDNN-HMM) architecture for acoustic model training to create our ASR system that is robust to environmental noise and to the variation of voice quality impacted by MCI. The LOTUS Thai speech corpus was incorporated into the training set to improve the model’s generalization. A preprocessing algorithm was implemented to reduce the background noise and improve the overall data quality before feeding data into the TDNN-HMM system for automatic word detection and language fluency score calculation. The results show that the TDNN-HMM model in combination with data augmentation using lattice-free maximum mutual information (LF-MMI) objective function provides a word error rate (WER) of 30.77%. To our knowledge, this is the first study to develop an ASR with Thai language support to automate the scoring system of MoCA’s language fluency assessment.
Open AccessUsing Automatic Speech Recognition to Assess Thai Speech Language Fluency in the Montreal Cognitive Assessment (MoCA)
The Montreal cognitive assessment (MoCA), a widely accepted screening tool for identifying patients with mild cognitive impairment (MCI), includes a language fluency test of verbal functioning; its scores are based on the number of unique correct words produced by the test taker. However, it is possible that unique words may be counted differently for various languages. This study focuses on Thai as a language that differs from English in terms of word combinations. We applied various automatic speech recognition (ASR) techniques to develop an assisted scoring system for the MoCA language fluency test with Thai language support. This was a challenge because Thai is a low-resource language for which domain-specific data are not publicly available, especially speech data from patients with MCIs. Furthermore, the great variety of pronunciation, intonation, tone, and accent of the patients, all of which might differ from healthy controls, bring more complexity to the model. We propose a hybrid time delay neural network hidden Markov model (TDNN-HMM) architecture for acoustic model training to create our ASR system that is robust to environmental noise and to the variation of voice quality impacted by MCI. The LOTUS Thai speech corpus was incorporated into the training set to improve the model’s generalization. A preprocessing algorithm was implemented to reduce the background noise and improve the overall data quality before feeding data into the TDNN-HMM system for automatic word detection and language fluency score calculation. The results show that the TDNN-HMM model in combination with data augmentation using lattice-free maximum mutual information (LF-MMI) objective function provides a word error rate (WER) of 30.77%. To our knowledge, this is the first study to develop an ASR with Thai language support to automate the scoring system of MoCA’s language fluency assessment.
- 4 February 2022
- Published by Cambridge University Press (CUP) in Natural Language Engineering
- Vol. 29 (2) , 254-286
Towards improving coherence and diversity of slogan generation
Previouswork in slogan generation focused on utilising slogan skeletons mined from existing slogans. While some generated slogans can be catchy, they are often not coherent with the company’s focus or style across their marketing communications because the skeletons are mined from other companies’ slogans. We propose a sequence-to-sequence (seq2seq) Transformer model to generate slogans from a brief company description. A naïve seq2seq model fine-tuned for slogan generation is prone to introducing false information. We use company name delexicalisation and entity masking to alleviate this problem and improve the generated slogans’ quality and truthfulness. Furthermore, we apply conditional training based on the first words’ part-of-speech tag to generate syntactically diverse slogans. Our best model achieved a ROUGE-1/-2/-L $\mathrm{F}_1$ score of 35.58/18.47/33.32. Besides, automatic and human evaluations indicate that our method generates significantly more factual, diverse and catchy slogans than strong long short-term memory and Transformer seq2seq baselines.- 4 February 2022
- Published by Cambridge University Press (CUP) in Natural Language Engineering
- Vol. 29 (2) , 254-286
Towards improving coherence and diversity of slogan generation
Previouswork in slogan generation focused on utilising slogan skeletons mined from existing slogans. While some generated slogans can be catchy, they are often not coherent with the company’s focus or style across their marketing communications because the skeletons are mined from other companies’ slogans. We propose a sequence-to-sequence (seq2seq) Transformer model to generate slogans from a brief company description. A naïve seq2seq model fine-tuned for slogan generation is prone to introducing false information. We use company name delexicalisation and entity masking to alleviate this problem and improve the generated slogans’ quality and truthfulness. Furthermore, we apply conditional training based on the first words’ part-of-speech tag to generate syntactically diverse slogans. Our best model achieved a ROUGE-1/-2/-L $\mathrm{F}_1$ score of 35.58/18.47/33.32. Besides, automatic and human evaluations indicate that our method generates significantly more factual, diverse and catchy slogans than strong long short-term memory and Transformer seq2seq baselines.- Open Access
Toward Improving Coherence and Diversity of Slogan Generation
Previous work in slogan generation focused on utilising slogan skeletons mined from existing slogans. While some generated slogans can be catchy, they are often not coherent with the company's focus or style across their marketing communications because the skeletons are mined from other companies' slogans. We propose a sequence-to-sequence (seq2seq) transformer model to generate slogans from a brief company description. A naive seq2seq model fine-tuned for slogan generation is prone to introducing false information. We use company name delexicalisation and entity masking to alleviate this problem and improve the generated slogans' quality and truthfulness. Furthermore, we apply conditional training based on the first words' POS tag to generate syntactically diverse slogans. Our best model achieved a ROUGE-1/-2/-L F1 score of 35.58/18.47/33.32. Besides, automatic and human evaluations indicate that our method generates significantly more factual, diverse and catchy slogans than strong LSTM and transformer seq2seq baselines.
Open AccessToward Improving Coherence and Diversity of Slogan Generation
Previous work in slogan generation focused on utilising slogan skeletons mined from existing slogans. While some generated slogans can be catchy, they are often not coherent with the company's focus or style across their marketing communications because the skeletons are mined from other companies' slogans. We propose a sequence-to-sequence (seq2seq) transformer model to generate slogans from a brief company description. A naive seq2seq model fine-tuned for slogan generation is prone to introducing false information. We use company name delexicalisation and entity masking to alleviate this problem and improve the generated slogans' quality and truthfulness. Furthermore, we apply conditional training based on the first words' POS tag to generate syntactically diverse slogans. Our best model achieved a ROUGE-1/-2/-L F1 score of 35.58/18.47/33.32. Besides, automatic and human evaluations indicate that our method generates significantly more factual, diverse and catchy slogans than strong LSTM and transformer seq2seq baselines.
- Open Access
Bootstrapping Large-Scale Fine-Grained Contextual Advertising Classifier from Wikipedia
Contextual advertising provides advertisers with the opportunity to target the context which is most relevant to their ads. However, its power cannot be fully utilized unless we can target the page content using fine-grained categories, e.g., "coupe" vs. "hatchback" instead of "automotive" vs. "sport". The widely used advertising content taxonomy (IAB taxonomy) consists of 23 coarse-grained categories and 355 fine-grained categories. With the large number of categories, it becomes very challenging either to collect training documents to build a supervised classification model, or to compose expert-written rules in a rule-based classification system. Besides, in fine-grained classification, different categories often overlap or co-occur, making it harder to classify accurately. In this work, we propose wiki2cat, a method to tackle the problem of large-scaled fine-grained text classification by tapping on Wikipedia category graph. The categories in IAB taxonomy are first mapped to category nodes in the graph. Then the label is propagated across the graph to obtain a list of labeled Wikipedia documents to induce text classifiers. The method is ideal for large-scale classification problems since it does not require any manually-labeled document or hand-curated rules or keywords. The proposed method is benchmarked with various learning-based and keyword-based baselines and yields competitive performance on both publicly available datasets and a new dataset containing more than 300 fine-grained categories.
Open AccessBootstrapping Large-Scale Fine-Grained Contextual Advertising Classifier from Wikipedia
Contextual advertising provides advertisers with the opportunity to target the context which is most relevant to their ads. However, its power cannot be fully utilized unless we can target the page content using fine-grained categories, e.g., "coupe" vs. "hatchback" instead of "automotive" vs. "sport". The widely used advertising content taxonomy (IAB taxonomy) consists of 23 coarse-grained categories and 355 fine-grained categories. With the large number of categories, it becomes very challenging either to collect training documents to build a supervised classification model, or to compose expert-written rules in a rule-based classification system. Besides, in fine-grained classification, different categories often overlap or co-occur, making it harder to classify accurately. In this work, we propose wiki2cat, a method to tackle the problem of large-scaled fine-grained text classification by tapping on Wikipedia category graph. The categories in IAB taxonomy are first mapped to category nodes in the graph. Then the label is propagated across the graph to obtain a list of labeled Wikipedia documents to induce text classifiers. The method is ideal for large-scale classification problems since it does not require any manually-labeled document or hand-curated rules or keywords. The proposed method is benchmarked with various learning-based and keyword-based baselines and yields competitive performance on both publicly available datasets and a new dataset containing more than 300 fine-grained categories.
- 17 June 2020
- Published by Cambridge University Press (CUP) in Natural Language Engineering
- Vol. 28 (1) , 39-69
Learning from noisy out-of-domain corpus using dataless classification
In real-world applications, text classification models often suffer from a lack of accurately labelled documents. The available labelled documents may also be out of domain, making the trained model not able to perform well in the target domain. In this work, we mitigate the data problem of text classification using a two-stage approach. First, we mine representative keywords from a noisy out-of-domain data set using statistical methods. We then apply a dataless classification method to learn from the automatically selected keywords and unlabelled in-domain data. The proposed approach outperformed various supervised learning and dataless classification baselines by a large margin. We evaluated different keyword selection methods intrinsically and extrinsically by measuring their impact on the dataless classification accuracy. Last but not least, we conducted an in-depth analysis of the behaviour of the classifier and explained why the proposed dataless classification method outperformed supervised learning counterparts.
- 17 June 2020
- Published by Cambridge University Press (CUP) in Natural Language Engineering
- Vol. 28 (1) , 39-69
Learning from noisy out-of-domain corpus using dataless classification
In real-world applications, text classification models often suffer from a lack of accurately labelled documents. The available labelled documents may also be out of domain, making the trained model not able to perform well in the target domain. In this work, we mitigate the data problem of text classification using a two-stage approach. First, we mine representative keywords from a noisy out-of-domain data set using statistical methods. We then apply a dataless classification method to learn from the automatically selected keywords and unlabelled in-domain data. The proposed approach outperformed various supervised learning and dataless classification baselines by a large margin. We evaluated different keyword selection methods intrinsically and extrinsically by measuring their impact on the dataless classification accuracy. Last but not least, we conducted an in-depth analysis of the behaviour of the classifier and explained why the proposed dataless classification method outperformed supervised learning counterparts.
- Open Access
Learning Only from Relevant Keywords and Unlabeled Documents
We consider a document classification problem where document labels are absent but only relevant keywords of a target class and unlabeled documents are given. Although heuristic methods based on pseudo-labeling have been considered, theoretical understanding of this problem has still been limited. Moreover, previous methods cannot easily incorporate well-developed techniques in supervised text classification. In this paper, we propose a theoretically guaranteed learning framework that is simple to implement and has flexible choices of models, e.g., linear models or neural networks. We demonstrate how to optimize the area under the receiver operating characteristic curve (AUC) effectively and also discuss how to adjust it to optimize other well-known evaluation metrics such as the accuracy and F1-measure. Finally, we show the effectiveness of our framework using benchmark datasets.
Open AccessLearning Only from Relevant Keywords and Unlabeled Documents
We consider a document classification problem where document labels are absent but only relevant keywords of a target class and unlabeled documents are given. Although heuristic methods based on pseudo-labeling have been considered, theoretical understanding of this problem has still been limited. Moreover, previous methods cannot easily incorporate well-developed techniques in supervised text classification. In this paper, we propose a theoretically guaranteed learning framework that is simple to implement and has flexible choices of models, e.g., linear models or neural networks. We demonstrate how to optimize the area under the receiver operating characteristic curve (AUC) effectively and also discuss how to adjust it to optimize other well-known evaluation metrics such as the accuracy and F1-measure. Finally, we show the effectiveness of our framework using benchmark datasets.