arXiv subjects

Jiyeon Kim

Publications and source records attributed to Jiyeon Kim.

34 records · Page 2Linked to original sources

arXiv2024

Data-driven grapheme-to-phoneme representations for a lexicon-free text-to-speech

Grapheme-to-Phoneme (G2P) is an essential first step in any modern, high-quality Text-to-Speech (TTS) system. Most of the current G2P systems rely on carefully hand-crafted lexicons developed by experts. This poses a two-fold problem. Firstly, the lexicons are generated using a fixed phoneme set, usually, ARPABET or IPA, which might not be the most optimal way to represent phonemes for all languages. Secondly, the man-hours required to produce such an expert lexicon are very high. In this paper, we eliminate both of these issues by using recent advances in self-supervised learning to obtain data-driven phoneme representations instead of fixed representations. We compare our lexicon-free approach against strong baselines that utilize a well-crafted lexicon. Furthermore, we show that our data-driven lexicon-free method performs as good or even marginally better than the conventional rule-based or lexicon-based neural G2Ps in terms of Mean Opinion Score (MOS) while using no prior language lexicon or phoneme set, i.e. no linguistic expertise.

cs.CL↗

arXiv2023

Prediction and control of two-dimensional decaying turbulence using generative adversarial networks

With the recent rapid developments in machine learning (ML), several attempts have been made to apply ML methods to various fluid dynamics problems. However, the feasibility of ML for predicting turbulence dynamics has not yet been explored in detail. In this study, PredictionNet, a data-driven ML framework based on generative adversarial networks (GANs), was developed to predict two-dimensional (2D) decaying turbulence. The developed prediction model accurately predicted turbulent fields at a finite lead time of up to half the Eulerian integral time scale. In addition to the high accuracy in pointwise metrics, various turbulence statistics, such as the probability density function, spatial correlation function, and enstrophy spectrum, were accurately captured by the employed GAN. Scale decomposition was used to interpret the predictability depending on the spatial scale, and the role of latent variables in the discriminator network was investigated. The good performance of the GAN in predicting small-scale turbulence is attributed to the scale-selection capability of the latent variable. Results also revealed that the recursive applications of the prediction model yielded better predictions than single predictions for large lead times. Furthermore, by utilizing PredictionNet as a surrogate model, a control model named ControlNet was developed to identify disturbance fields that drive the time evolution of the flow field in the direction that optimises the specified objective function. Therefore, an illustrative example in which the evolution of 2D turbulence can be predicted within a finite time horizon and controlled using a GAN-based deep neural network is presented.

physics.flu-dyn↗

arXiv2023

Improving Out-of-Distribution Detection in Echocardiographic View Classication through Enhancing Semantic Features

In echocardiographic view classification, accurately detecting out-of-distribution (OOD) data is essential but challenging, especially given the subtle differences between in-distribution and OOD data. While conventional OOD detection methods, such as Mahalanobis distance (MD) are effective in far-OOD scenarios with clear distinctions between distributions, they struggle to discern the less obvious variations characteristic of echocardiographic data. In this study, we introduce a novel use of label smoothing to enhance semantic feature representation in echocardiographic images, demonstrating that these enriched semantic features are key for significantly improving near-OOD instance detection. By combining label smoothing with MD-based OOD detection, we establish a new benchmark for accuracy in echocardiographic OOD detection.

eess.SP↗

arXiv2023

A Unified Approach for Comprehensive Analysis of Various Spectral and Tissue Doppler Echocardiography

Doppler echocardiography offers critical insights into cardiac function and phases by quantifying blood flow velocities and evaluating myocardial motion. However, previous methods for automating Doppler analysis, ranging from initial signal processing techniques to advanced deep learning approaches, have been constrained by their reliance on electrocardiogram (ECG) data and their inability to process Doppler views collectively. We introduce a novel unified framework using a convolutional neural network for comprehensive analysis of spectral and tissue Doppler echocardiography images that combines automatic measurements and end-diastole (ED) detection into a singular method. The network automatically recognizes key features across various Doppler views, with novel Doppler shape embedding and anti-aliasing modules enhancing interpretation and ensuring consistent analysis. Empirical results indicate a consistent outperformance in performance metrics, including dice similarity coefficients (DSC) and intersection over union (IoU). The proposed framework demonstrates strong agreement with clinicians in Doppler automatic measurements and competitive performance in ED detection.

eess.IV↗

arXiv2022

First-principles study on Small Polaron and Li diffusion in layered LiCoO2

Li-ion conductivity is one of the essential properties that determine the performance of cathode materials for Li-ion batteries. Here, using the density functional theory, we investigate the polaron stability and its effect on the Li-ion diffusion in layered LiCoO2 with different magnetic orderings. The localized Co4+ polaron appears in the magnetic configurations and sets the Li-diffusion barrier of ~0.34 eV. The polaron also migrates in the opposite direction to the Li-diffusion direction. On the other hand, the polaron does not form in the non-magnetic structure, and the Li diffusion barrier without the polaron is 0.21 eV. Although the existence of the polaron increases the diffusion barrier, the magnetically ordered structures are more energetically stable during the migration than the non-magnetic case. Thus, our work advocates the hole polaron migration scenario for Li-ion diffusion. Moreover, we demonstrate that the strong electron correlation of Co ions plays an essential role in stabilizing the Co4+ polaron.

cond-mat.mtrl-sci↗

arXiv2022

Stability and electronic properties of layered NaMnO2 using the SCAN(+U)

Considering electron correlation appropriately is important to predict the properties of layered transition metal oxides, which have drawn a lot of attention as cathode materials for sodium-ion batteries. Here, we explore the phonon and electronic properties of layered NaMnO2 using the recently developed strongly constrained and appropriately normed (SCAN) functional. We also introduce the Coulomb interaction U to find an accurate description of Mn 3d orbitals. The phonon dispersion curves show the structural stability with the SCAN, which is consistent with prior experimental stability at high Na concentrations. On the other hand, imaginary phonon frequencies were observed by applying U, which indicates structural instability. Namely, SCAN properly describes the phonon properties of layered NaMnO2 whereas SCAN+U does not. We further explore the Jahn-Teller (J-T) stability and magnitude of J-T distortion depending on U and find that SCAN results are consistent with Perdew-Burke-Ernzerhof functional, PBE+U results. Our results suggest that SCAN itself properly describes the physical properties of NaMnO2 without adding U explicitly.

cond-mat.mtrl-sci↗

arXiv2022

Deep reinforcement learning for large-eddy simulation modeling in wall-bounded turbulence

The development of a reliable subgrid-scale (SGS) model for large-eddy simulation (LES) is of great importance for many scientific and engineering applications. Recently, deep learning approaches have been tested for this purpose using high-fidelity data such as direct numerical simulation (DNS) in a supervised learning process. However, such data are generally not available in practice. Deep reinforcement learning (DRL) using only limited target statistics can be an alternative algorithm in which the training and testing of the model are conducted in the same LES environment. The DRL of turbulence modeling remains challenging owing to its chaotic nature, high dimensionality of the action space, and large computational cost. In the present study, we propose a physics-constrained DRL framework that can develop a deep neural network (DNN)-based SGS model for the LES of turbulent channel flow. The DRL models that produce the SGS stress were trained based on the local gradient of the filtered velocities. The developed SGS model automatically satisfies the reflectional invariance and wall boundary conditions without an extra training process so that DRL can quickly find the optimal policy. Furthermore, direct accumulation of reward, spatially and temporally correlated exploration, and the pre-training process are applied for the efficient and effective learning. In various environments, our DRL could discover SGS models that produce the viscous and Reynolds stress statistics perfectly consistent with the filtered DNS. By comparing various statistics obtained by the trained models and conventional SGS models, we present a possible interpretation of better performance of the DRL model.

physics.flu-dyn↗

arXiv2022

Phonon study of Jahn-Teller distortion and phase stability in NaMnO$_2$ for sodium-ion batteries

Cathode materials undergo various phase transitions during the charge/discharge process, and the structural transitions significantly affect the battery performance. Although phonon properties can provide a direct clue for structural stability and transitions, it has been less explored in sodium cathode materials. Here, using the first-principles calculations, we investigate phonon and electronic properties of various layered NaMnO$_2$ materials, especially focusing on the dependency of the Jahn-Teller distortion of Mn$^{3+}$. The phonon dispersion curves show that the O$'$3 and P$'$2 structures with the Jahn-Teller distortion are dynamically stable in contrast to undistorted O3 and P2 structures. The structural instability of O3 and P2 structures is directly observed from the imaginary phonon frequencies, as so-called phonon soft modes, whose corresponding displacements are from O atoms distorting along the local Mn-O bond direction in the MnO$_6$ octahedra. This is consistent with the experimental stability and a structural transition with the Jahn-Teller distortion at the high Na concentration. Furthermore, the orbital-decomposed density of states presents the orbital redistribution by the Jahn-Teller distortion such as $e_g$-band splitting, and the stability of O$'$3 and P$'$2 is not sensitive to the electron-electron correlation. Our results demonstrate the importance of phonon analysis to further understand the structural stability and phase transitions in cathode materials.

cond-mat.mtrl-sci↗

arXiv2021

A comparison of streaming models and data augmentation methods for robust speech recognition

In this paper, we present a comparative study on the robustness of two different online streaming speech recognition models: Monotonic Chunkwise Attention (MoChA) and Recurrent Neural Network-Transducer (RNN-T). We explore three recently proposed data augmentation techniques, namely, multi-conditioned training using an acoustic simulator, Vocal Tract Length Perturbation (VTLP) for speaker variability, and SpecAugment. Experimental results show that unidirectional models are in general more sensitive to noisy examples in the training set. It is observed that the final performance of the model depends on the proportion of training examples processed by data augmentation techniques. MoChA models generally perform better than RNN-T models. However, we observe that training of MoChA models seems to be more sensitive to various factors such as the characteristics of training sets and the incorporation of additional augmentations techniques. On the other hand, RNN-T models perform better than MoChA models in terms of latency, inference time, and the stability of training. Additionally, RNN-T models are generally more robust against noise and reverberation. All these advantages make RNN-T models a better choice for streaming on-device speech recognition compared to MoChA models.

eess.AS↗

arXiv2021

Semi-supervised transfer learning for language expansion of end-to-end speech recognition models to low-resource languages

In this paper, we propose a three-stage training methodology to improve the speech recognition accuracy of low-resource languages. We explore and propose an effective combination of techniques such as transfer learning, encoder freezing, data augmentation using Text-To-Speech (TTS), and Semi-Supervised Learning (SSL). To improve the accuracy of a low-resource Italian ASR, we leverage a well-trained English model, unlabeled text corpus, and unlabeled audio corpus using transfer learning, TTS augmentation, and SSL respectively. In the first stage, we use transfer learning from a well-trained English model. This primarily helps in learning the acoustic information from a resource-rich language. This stage achieves around 24% relative Word Error Rate (WER) reduction over the baseline. In stage two, We utilize unlabeled text data via TTS data-augmentation to incorporate language information into the model. We also explore freezing the acoustic encoder at this stage. TTS data augmentation helps us further reduce the WER by ~ 21% relatively. Finally, In stage three we reduce the WER by another 4% relative by using SSL from unlabeled audio data. Overall, our two-pass speech recognition system with a Monotonic Chunkwise Attention (MoChA) in the first pass and a full-attention in the second pass achieves a WER reduction of ~ 42% relative to the baseline.

eess.AS↗

arXiv2021

A review of on-device fully neural end-to-end automatic speech recognition algorithms

In this paper, we review various end-to-end automatic speech recognition algorithms and their optimization techniques for on-device applications. Conventional speech recognition systems comprise a large number of discrete components such as an acoustic model, a language model, a pronunciation model, a text-normalizer, an inverse-text normalizer, a decoder based on a Weighted Finite State Transducer (WFST), and so on. To obtain sufficiently high speech recognition accuracy with such conventional speech recognition systems, a very large language model (up to 100 GB) is usually needed. Hence, the corresponding WFST size becomes enormous, which prohibits their on-device implementation. Recently, fully neural network end-to-end speech recognition algorithms have been proposed. Examples include speech recognition systems based on Connectionist Temporal Classification (CTC), Recurrent Neural Network Transducer (RNN-T), Attention-based Encoder-Decoder models (AED), Monotonic Chunk-wise Attention (MoChA), transformer-based speech recognition systems, and so on. These fully neural network-based systems require much smaller memory footprints compared to conventional algorithms, therefore their on-device implementation has become feasible. In this paper, we review such end-to-end speech recognition models. We extensively discuss their structures, performance, and advantages compared to conventional algorithms.

cs.LG↗

arXiv2019

end-to-end training of a large vocabulary end-to-end speech recognition system

In this paper, we present an end-to-end training framework for building state-of-the-art end-to-end speech recognition systems. Our training system utilizes a cluster of Central Processing Units(CPUs) and Graphics Processing Units (GPUs). The entire data reading, large scale data augmentation, neural network parameter updates are all performed "on-the-fly". We use vocal tract length perturbation [1] and an acoustic simulator [2] for data augmentation. The processed features and labels are sent to the GPU cluster. The Horovod allreduce approach is employed to train neural network parameters. We evaluated the effectiveness of our system on the standard Librispeech corpus [3] and the 10,000-hr anonymized Bixby English dataset. Our end-to-end speech recognition system built using this training infrastructure showed a 2.44 % WER on test-clean of the LibriSpeech test set after applying shallow fusion with a Transformer language model (LM). For the proprietary English Bixby open domain test set, we obtained a WER of 7.92 % using a Bidirectional Full Attention (BFA) end-to-end model after applying shallow fusion with an RNN-LM. When the monotonic chunckwise attention (MoCha) based approach is employed for streaming speech recognition, we obtained a WER of 9.95 % on the same Bixby open domain test set.

eess.AS↗

arXiv2019

Deep ensemble network with explicit complementary model for accuracy-balanced classification

The average accuracy is one of major evaluation metrics for classification systems, while the accuracy deviation is another important performance metric used to evaluate various deep neural networks. In this paper, we present a new ensemble-like fast deep neural network, Harmony, that can reduce the accuracy deviation among categories without degrading overall average accuracy. Harmony consists of three sub-models, namely, Target model, Complementary model, and Conductor model. In Harmony, an object is classified by using either Target model or Complementary model. Target model is a conventional classification network for general categories, while Complementary model is a classification network especially for weak categories that are inaccurately classified by Target model. Conductor model is used to select one of two models. Experimental results demonstrate that Harmony accurately classifies categories, while it reduces the accuracy deviation among the categories.

cs.CV↗

arXiv2016

Robust singlet dimers with fragile ordering in two-dimensional honeycomb lattice of Li$_2$RuO$_3$

When an electronic system has strong correlations and a large spin-orbit interaction, it often exhibits a plethora of mutually competing quantum phases. How a particular quantum ground state is selected out of several possibilities is a very interesting question. However, equally fascinating is how such a quantum entangled state breaks up due to perturbation. This important question has relevance in very diverse fields of science from strongly correlated electron physics to quantum information. Here we report that a quantum entangled dimerized state or valence bond crystal (VBC) phase of Li2RuO3 shows nontrivial doping dependence as we perturb the Ru honeycomb lattice by replacing Ru with Li. Through extensive experimental studies, we demonstrate that the VBC phase melts into a valence bond liquid phase of the RVB (resonating valence bond) type. This system offers an interesting playground where one can test and refine our current understanding of the quantum competing phases in a single compound.

cond-mat.str-el↗

arXiv2014

Experimental percolation studies of two-dimensional honeycomb lattice: Li2Mn1-xTixO3

Li2MnO3 with a S=3/2 two-dimensional Mn honeycomb lattice has a Neel-type antiferromagnetic transition at TN=36 K with a broad maximum in the magnetic susceptibility at TM=48 K. We have investigated site percolation effects by replacing Mn with nonmagnetic Ti, and completed a full phase diagram of Li2Mn1-xTixO3 solid solution systems to find that the antiferromagnetic transition is suppressed continuously without a clear sign of changes in the Neel-type antiferromagnetic structure. The magnetic ordering eventually disappears at a critical concentration of xc=0.7. This experimental observation is consistent with percolation theories for a honeycomb lattice when one considers up to 3rd nearest-neighbor interactions. This study highlights the importance of interaction beyond nearest neighbors even for Mn element with relative localized 3d electrons in the honeycomb lattice.

cond-mat.str-el↗

arXiv2012

Physical properties of transparent perovskite oxides (Ba,La)SnO3 with high electrical mobility at room temperature

Transparent electronic materials are increasingly in demand for a variety of optoelectronic applications. BaSnO3 is a semiconducting oxide with a large band gap of more than 3.1 eV. Recently, we discovered that La doped BaSnO3 exhibits unusually high electrical mobility of 320 cm^2(Vs)^-1 at room temperature and superior thermal stability at high temperatures [H. J. Kim et al. Appl. Phys. Express. 5, 061102 (2012)]. Following that work, we report various physical properties of (Ba,La)SnO3 single crystals and films including temperature-dependent transport and phonon properties, optical properties and first-principles calculations. We find that almost doping-independent mobility of 200-300 cm^2(Vs)^-1 is realized in the single crystals in a broad doping range from 1.0x10^19 to 4.0x10^20 cm^-3. Moreover, the conductivity of ~10^4 ohm^-1cm^-1 reached at the latter carrier density is comparable to the highest value. We attribute the high mobility to several physical properties of (Ba,La)SnO3: a small effective mass coming from the ideal Sn-O-Sn bonding, small disorder effects due to the doping away from the SnO2 conduction channel, and reduced carrier scattering due to the high dielectric constant. The observation of a reduced mobility of ~70 cm^2(Vs)^-1 in the film is mainly attributed to additional carrier-scatterings which are presumably created by the lattice mismatch between the substrate SrTiO3 and (Ba,La)SnO3. The main optical gap of (Ba,La)SnO3 single crystals remained at about 3.33 eV and the in-gap states only slightly increased, thus maintaining optical transparency in the visible region. Based on these, we suggest that the doped BaSnO3 system holds great potential for realizing all perovskite-based, transparent high-frequency high-power functional devices as well as highly mobile two-dimensional electron gas via interface control of heterostructured films.

cond-mat.mtrl-sci↗