• ETRI Journal
• /
• 제45권4호
• /
• pp.678-689
• /
• 2023

This paper describes a novel end-to-end deep generative model-based speaker recognition system using prosodic features. The usefulness of variational autoencoders (VAE) in learning the speaker-specific prosody representations for the speaker recognition task is examined herein for the first time. The speech signal is first automatically segmented into syllable-like units using vowel onset points (VOP) and energy valleys. Prosodic features, such as the dynamics of duration, energy, and fundamental frequency (F₀), are then extracted at the syllable level and used to train/adapt a speaker-dependent VAE from a universal VAE. The initial comparative studies on VAEs and traditional autoencoders (AE) suggest that the former can efficiently learn speaker representations. Investigations on the impact of gender information in speaker recognition also point out that gender-dependent impostor banks lead to higher accuracies. Finally, the evaluation on the NIST SRE 2010 dataset demonstrates the usefulness of the proposed approach for speaker recognition.

• Journal of information and communication convergence engineering
• /
• 제22권1호
• /
• pp.56-63
• /
• 2024

Despite the rapid strides in content-based image retrieval, a notable disparity persists between the visual features of images and the semantic features discerned by humans. Hence, image retrieval based on the association of semantic similarities recognized by humans with visual similarities is a difficult task for most image-retrieval systems. Our study endeavors to bridge this gap by refining image semantics, aligning them more closely with human perception. Deep learning techniques are used to semantically classify images and retrieve those that are semantically similar to personalized images. Moreover, we introduce a keyword-based image retrieval, enabling automatic labeling of images in mobile environments. The proposed approach can improve the performance of a mobile device with limited resources and bandwidth by performing retrieval based on the visual features and keywords of the image on the mobile device.

• 대한기관식도과학회지
• /
• 제2권2호
• /
• pp.253-257
• /
• 1996

Deep neck infections were flirty common and a source of considerable morbidity and mortality. Although the advent of antibiotics has reduced the overall number of deep neck infections, they still occur in the general population. There are several new groups of patients at risk for deep neck infections, such as immunocompromised individuals, those with underlying diseases. Prevention of the severe sequale that may be associated with deep neck infections- mediastinitis, airway obstruction, carotid artery hemorrhage, aspiration pneumonia, septicemia - requires a knowledge of various portals of entry for infection, the presenting sign and symptoms, the possible microbiologic features, appropriate laboratory and radiologic workups, therapeutic techniques, and the ongoing medical management. A prompt diagnosis and institution of therapy will shorten the course of required treatment and reduce morbility and mortility. The authors have experienced one case of acute mediastinitis in deep neck infection patient with diabetes mellitus.

• 융합보안논문지
• /
• 제23권4호
• /
• pp.61-70
• /
• 2023

클러스터링은 데이터의 정답값(실제값)이 없는 데이터를 기반으로 데이터의 특징벡터의 거리 기반 등으로 군집화를 하는 비지도학습 방법이다. 이 방법은 이미지, 텍스트, 음성 등 다양한 데이터에 대해서 라벨링이 없이 적용할 수 있다는 장점이 있다. 기존 클러스터링을 하기 위해 차원축소 기법을 적용하거나 특정 특징만을 추출하여 군집화하는 방법이 적용되었다. 하지만 딥러닝 기반 모델이 발전하면서 입력 데이터를 잠재 벡터로 표현하는 오토인코더, 생성 적대적 네트워크 등을 통해서 딥 클러스터링의 기술이 연구가 되고 있다. 본 연구에서, 딥러닝 기반의 딥 클러스터링 기법을 제안하였다. 이 방법에서 오토인코더를 이용하여 입력 데이터를 잠재 벡터로 변환하고 이 잠재 벡터를 클러스터 구조에 맞게 벡터 공간을 구성 및 k-평균 클러스터링을 하였다. 실험 환경으로 pytorch 머신러닝 라이브러리를 이용하여 데이터셋으로 MNIST와 Fashion-MNIST을 적용하였다. 모델로는 컨볼루션 신경망 기반인 오토인코더 모델을 사용하였다. 실험결과로 k가 10일 때, MNIST에 대해서 89.42% 정확도를 가졌으며 Fashion-MNIST에 대해서 56.64% 정확도를 가진다.

• ETRI Journal
• /
• 제42권1호
• /
• pp.90-100
• /
• 2020

Named entity recognition (NER) continues to be an important task in natural language processing because it is featured as a subtask and/or subproblem in information extraction and machine translation. In Urdu language processing, it is a very difficult task. This paper proposes various deep recurrent neural network (DRNN) learning models with word embedding. Experimental results demonstrate that they improve upon current state-of-the-art NER approaches for Urdu. The DRRN models evaluated include forward and bidirectional extensions of the long short-term memory and back propagation through time approaches. The proposed models consider both language-dependent features, such as part-of-speech tags, and language-independent features, such as the "context windows" of words. The effectiveness of the DRNN models with word embedding for NER in Urdu is demonstrated using three datasets. The results reveal that the proposed approach significantly outperforms previous conditional random field and artificial neural network approaches. The best f-measure values achieved on the three benchmark datasets using the proposed deep learning approaches are 81.1%, 79.94%, and 63.21%, respectively.

• 한국멀티미디어학회논문지
• /
• 제23권8호
• /
• pp.965-976
• /
• 2020

We present in this work a segmentation method of E. coli bacterial images generated via phase contrast microscopy using a deep learning based hybrid feature generation. Unlike conventional machine learning methods that use the hand-crafted features, we adopt the denoising autoencoder in order to generate a precise and accurate representation of the pixels. We first construct a hybrid vector that combines original image, difference of Gaussians and image gradients. The created hybrid features are then given to a deep autoencoder that learns the pixels' internal dependencies and the cells' shape and boundary information. The latent representations learned by the autoencoder are used as the inputs of a softmax classification layer and the direct outputs from the classifier represent the coarse segmentation mask. Finally, the classifier's outputs are used as prior information for a graph partitioning based fine segmentation. We demonstrate that the proposed hybrid vector representation manages to preserve the global shape and boundary information of the cells, allowing to retrieve the majority of the cellular patterns without the need of any post-processing.

• International Journal of Contents
• /
• 제15권4호
• /
• pp.59-64
• /
• 2019

In recent years, emotion recognition has been an interesting and challenging topic. Compared to facial expressions and speech modality, gesture-based emotion recognition has not received much attention with only a few efforts using traditional hand-crafted methods. These approaches require major computational costs and do not offer many opportunities for improvement as most of the science community is conducting their research based on the deep learning technique. In this paper, we propose an end-to-end deep learning approach for classifying emotions based on bodily gestures. In particular, the informative keyframes are first extracted from raw videos as input for the 3D-CNN deep network. The 3D-CNN exploits the short-term spatiotemporal information of gesture features from selected keyframes, and the convolutional LSTM networks learn the long-term feature from the features results of 3D-CNN. The experimental results on the FABO dataset exceed most of the traditional methods results and achieve state-of-the-art results for the deep learning-based technique for gesture-based emotion recognition.

• Geomechanics and Engineering
• /
• 제33권2호
• /
• pp.203-209
• /
• 2023

One major advantage of ground penetrating radar (GPR) over other field test methods is its ability to obtain subsurface images of roads in an efficient and non-intrusive manner. Not only can the strata of pavement structure be retrieved from the GPR scan images, but also various irregularities, such as cracks and internal cavities. This article introduces a deep learning-based approach, focusing on detecting subsurface cracks by recognizing their distinctive hyperbolic signatures in the GPR scan images. Given the limited road sections that contain target features, two data augmentation methods, i.e., feature insertion and generation, are implemented, resulting in 9,174 GPR scan images. One of the most popular real-time object detection models, You Only Learn One Representation (YOLOR), is trained for detecting the target features for two types of subsurface cracks: bottom cracks and full cracks from the GPR scan images. The former represents partial cracks initiated from the bottom of the asphalt layer or base layers, while the latter includes extended cracks that penetrate these layers. Our experiments show the test average precisions of 0.769, 0.803 and 0.735 for all cracks, bottom cracks, and full cracks, respectively. This demonstrates the practicality of deep learning-based methods in detecting subsurface cracks from GPR scan images.

• International journal of advanced smart convergence
• /
• 제12권4호
• /
• pp.142-146
• /
• 2023

We present a method for generating 3D structures and rendering objects by combining VAE (Variational Autoencoder) and GAN (Generative Adversarial Network). This approach focuses on generating and rendering 3D models with improved quality using residual learning as the learning method for the encoder. We deep stack the encoder layers to accurately reflect the features of the image and apply residual blocks to solve the problems of deep layers to improve the encoder performance. This solves the problems of gradient vanishing and exploding, which are problems when constructing a deep neural network, and creates a 3D model of improved quality. To accurately extract image features, we construct deep layers of the encoder model and apply the residual function to learning to model with more detailed information. The generated model has more detailed voxels for more accurate representation, is rendered by adding materials and lighting, and is finally converted into a mesh model. 3D models have excellent visual quality and accuracy, making them useful in various fields such as virtual reality, game development, and metaverse.

• 인터넷정보학회논문지
• /
• 제25권1호
• /
• pp.49-56
• /
• 2024

There have been emerging many use-cases applying recommendation systems especially in online platform. Although the performance of recommendation systems is affected by a variety of factors, selecting appropriate features is difficult since most of recommendation systems have sparse data. Conventional matrix factorization (MF) method is a basic way to handle with problems in the recommendation systems. However, the MF based scheme cannot reflect non-linearity characteristics well. As deep learning technology has been attracted widely, a deep neural network (DNN) framework based collaborative filtering (CF) was introduced to complement the non-linearity issue. However, there is still a problem related to feature embedding for use as input to the DNN. In this paper, we propose an effective method using singular value decomposition (SVD) based feature embedding for improving the DNN performance of recommendation algorithms. We evaluate the performance of recommendation systems using MovieLens dataset and show the proposed scheme outperforms the existing methods. Moreover, we analyze the performance according to the number of latent features in the proposed algorithm. We expect that the proposed scheme can be applied to the generalized recommendation systems.