A5101455200- Speech Recognition and Synthesis
- Cancer, Stress, Anesthesia, and Immune Response
- Music and Audio Processing
- Speech and Audio Processing
- Speech and dialogue systems
- Topic Modeling
- Medical Imaging and Analysis
- Anesthesia and Neurotoxicity Research
- Acute Ischemic Stroke Management
- Natural Language Processing Techniques
- Photoacoustic and Ultrasonic Imaging
- Receptor Mechanisms and Signaling
- Gastrointestinal Bleeding Diagnosis and Treatment
- Advanced X-ray and CT Imaging
- Context-Aware Activity Recognition Systems
- Hand Gesture Recognition Systems
- Dental Radiography and Imaging
- Sinusitis and nasal conditions
- Intensive Care Unit Cognitive Disorders
- Uterine Myomas and Treatments
- Chemokine receptors and signaling
- Colorectal Cancer Screening and Detection
- Musculoskeletal synovial abnormalities and treatments
- Head and Neck Surgical Oncology
- Phonetics and Phonology Research
Pohang University of Science and Technology
2023-2024
Seoul National University Bundang Hospital
2009-2024
Korea Post
2023-2024
Ewha Womans University
2019-2023
Ewha Womans University Medical Center
2022
Accurate image interpretation of Waters’ and Caldwell view radiographs used for sinusitis screening is challenging. Therefore, we developed a deep learning algorithm diagnosing frontal, ethmoid, maxillary on both views. The datasets were selected the training validation set (n = 1403, sinusitis% 34.3%) test 132, 29.5%) by temporal separation. can simultaneously detect classify each paranasal sinus using views without manual cropping. Single- multi-view models compared. Our proposed...
Background The optimal first-line mechanical thrombectomy (MT) method in cancer-related stroke (CRS) patients with emergent large vessel occlusion (ELVO) remains largely unknown. aim of this study is to evaluate the efficacy and safety between contact aspiration (CA) stent retriever (SR) CRS patients. Methods Sixty-two ELVO, who underwent MT January 2013 October 2019 at our institution, were retrospectively analyzed. Patients divided into two groups based on compared: CA group (n=28), which...
Psoriasis is a chronic skin inflammation caused by dysfunctional immune system, which causes systemic in various organs and tissues. Due to the risk of recurrence psoriasis, it important identify critical targets pathogenesis psoriasis develop targeted therapeutics. Dimerized translationally controlled tumor protein (dTCTP) promotes cell activation as pro-inflammatory cytokine plays role developing allergic diseases such asthma rhinitis. Here, we sought explore whether dTCTP its inhibition...
Abstract Background/Objectives Translationally controlled tumor protein (TCTP) exhibits numerous biological functions. It has been shown to be involved in the regulation of glucose. However, its specific role metabolism not yet clearly elucidated. Here, we aimed assess effect TCTP overexpression on metabolic tissues and systemic energy metabolism. Subjects/Methods We investigated whether can ameliorate imbalance that causes obesity using TCTP-overexpressing transgenic (TCTP TG) mice. The...
Translationally controlled tumor protein (TCTP), a highly conserved present in most eukaryotes, is involved numerous biological processes. Only the dimeric form of TCTP (dTCTP) formed during inflammatory conditions exhibits cytokine-like activity. Therefore, dTCTP considered as therapeutic target for allergic diseases. Because monomeric (mTCTP) and share high topological similarity, we hypothesized that small molecules interacting with mTCTP would also bind to interfere dTCTP-based cellular...
Because most of the capsule-endoscopic images contain normal mucous membranes, physicians spend their reading time observing areas. Thus, a significant reduction in would be possible if only portion image frame for which particular lesion is suspected can read intensively. This study aims to develop deep convolutional neural-network-based model capable automatically detecting lesions small bowel. The proposed consists two neural networks parallel, each takes RGB and CIELab color spaces,...
Zero-shot multi-speaker TTS aims to synthesize speech with the voice of a chosen target speaker without any fine-tuning. Prevailing methods, however, encounter limitations at adapting new speakers out-of-domain settings, primarily due inadequate disentanglement and content leakage. To overcome these constraints, we propose an innovative negation feature learning paradigm that models decoupled attributes as deviations from complete audio representation by utilizing subtraction operation. By...
Zero-shot multi-speaker TTS aims to synthesize speech with the voice of a chosen target speaker without any fine-tuning. Prevailing methods, however, encounter limitations at adapting new speakers out-of-domain settings, primarily due inadequate disentanglement and content leakage. To overcome these constraints, we propose an innovative negation feature learning paradigm that models decoupled attributes as deviations from complete audio representation by utilizing subtraction operation. By...
Research on hate speech has predominantly revolved around detection and interpretation from textual inputs, leaving verbal content largely unexplored. While there been limited exploration into within acoustic the aspect of interpretability overlooked. Therefore, we introduce a new task explainable audio detection. Specifically, aim to identify precise time intervals, referred as frame-level rationales, which serve evidence for classification. Towards this end, propose two different...
Contemporary neural speech synthesis models have indeed demonstrated remarkable proficiency in synthetic generation as they attained a level of quality comparable to that human-produced speech. Nevertheless, it is important note these achievements predominantly been verified within the context high-resource languages such English. Furthermore, Tacotron and FastSpeech variants show substantial pausing errors when applied Korean language, which affects perception naturalness. In order address...
The difficulty of acquiring abundant, high-quality data, especially in multi-lingual contexts, has sparked interest addressing low-resource scenarios. Moreover, current literature rely on fixed expressions from language IDs, which results the inadequate learning representations, and failure to generate speech unseen languages. To address these challenges, we propose a novel method that directly extracts linguistic features audio input while effectively filtering out miscellaneous acoustic...
This study describes the effects of translationally controlled tumor protein (TCTP) on mice with memory impairment caused by scopolamine (SCO) administration. Specifically, functions and expression levels hippocampal synaptic proteins in 7- to 12-month-old SCO-treated wild-type (WT-SCO) were compared those TCTP-overexpressing (TG) TCTP knocked-down (KD) similarly treated SCO. Passive-avoidance tasks performed WT, TG, KD for four weeks after intraperitoneal injection SCO or saline followed an...
Dialogue state tracking plays a crucial role in extracting information task-oriented dialogue systems. However, preceding research are limited to textual modalities, primarily due the shortage of authentic human audio datasets. We address this by investigating synthetic data for audio-based DST. To end, we develop cascading and end-to-end models, train them with our dataset, test on actual speech data. facilitate evaluation tailored introduce novel PhonemeF1 capture pronunciation similarity....
Dialogue state tracking plays a crucial role in extracting information task-oriented dialogue systems. However, preceding research are limited to textual modalities, primarily due the shortage of authentic human audio datasets. We address this by investigating synthetic data for audio-based DST. To end, we develop cascading and end-to-end models, train them with our dataset, test on actual speech data. facilitate evaluation tailored introduce novel PhonemeF1 capture pronunciation similarity....
Abstract Background: The diverse roles of ubiquitously present translationally controlled tumor protein (TCTP) have been well delineated in several organs, but its possible function the brain, especially with regard to memory function, has not received much attention. This study describes effects TCTP on mice impaired by scopolamine (SCO) administration. Specifically, and synaptic functions 7- 12-month-old SCO-treated wild (WT) were compared those TCTP-overexpressing (TG) knocked down (KD)...
Acute ischemic stroke (AIS) is not only a common cause of disability but also leading mortality worldwide. Recent studies have shown that the collateral status could play vital role in assessing AIS and determining treatment options for patients. Herein, we propose joint regression ordinal learning approach AIS, built upon 3-D deep convolutional neural networks, facilitates an automated objective imaging from dynamic susceptibility contrast-enhanced magnetic resonance perfusion (DSC-MRP)....