• Phonetics and Speech Sciences
• /
• v.15 no.3
• /
• pp.83-88
• /
• 2023

This study investigates the fine-tuning of large-scale Automatic Speech Recognition (ASR) models, specifically OpenAI's Whisper model, for domain-specific applications using the KsponSpeech dataset. The primary research questions address the effectiveness of targeted lexical item emphasis during fine-tuning, its impact on domain-specific performance, and whether the fine-tuned model can maintain generalization capabilities across different languages and environments. Experiments were conducted using two fine-tuning datasets: Set A, a small subset emphasizing specific lexical items, and Set B, consisting of the entire KsponSpeech dataset. Results showed that fine-tuning with targeted lexical items increased recognition accuracy and improved domain-specific performance, with generalization capabilities maintained when fine-tuned with a smaller dataset. For noisier environments, a trade-off between specificity and generalization capabilities was observed. This study highlights the potential of fine-tuning using minimal domain-specific data to achieve satisfactory results, emphasizing the importance of balancing specialization and generalization for ASR models. Future research could explore different fine-tuning strategies and novel technologies such as prompting to further enhance large-scale ASR models' domain-specific performance.

• Journal of Korean Society for Atmospheric Environment
• /
• v.13 no.2
• /
• pp.137-145
• /
• 1997

To predict the effects of air pollutant in the coastal region, we have developed the air pollutant model, the reaction model and the deposition of NO, $NO_2, and O_3$. And the numerical model of air pollutant concentration employed the nested technique to calculate with the higher resolution for the area. The nested technique used two grid systems, one for the large scale calculating region with the coarse mesh grid (CMG) and the other for the small scale region with the fine grid (FMG). In other to prove the validity of the simulation model the calculations were conducted for the present situation. The results of them reasonably agree with the observed data and proved the validity of the model.

• Wind and Structures
• /
• v.20 no.6
• /
• pp.719-737
• /
• 2015

Full scale measurement on the structural dynamic characteristics and Vortex-induced Vibrations (VIV) of a long-span suspension bridge with a central span of 1650 m were conducted. Different Finite Element (FE) modeling principles for the separated twin-box girder were compared and evaluated with the field vibration test results, and the double-spine model was determined to be the best simulation model, but certain modification still needs to be made which will affect the basic modeling parameters and the dynamic response prediction values of corresponding wind tunnel tests. Based on the FE modal analysis results, small-scaled and large-scaled sectional model tests were both carried out to investigate the VIV responses, and probable Reynolds Number effects or scale effect on VIV responses were presented. Based on the observed VIV modes in the field measurement, the VIV results obtained from sectional model tests were converted into those of the three-dimensional (3D) full-scale bridge and subsequently compared with field measurement results. It is indicated that the large-scaled sectional model test can probably provide a reasonable and effective prediction on VIV response.

• Journal of the Society of Disaster Information
• /
• v.4 no.2
• /
• pp.10-27
• /
• 2008

When the ground is excavated adjacent to the existing tunnel, which is loaded by the surcharge on the ground surface, the tunnel stability would be very sensitive to the deformation of the ground induced by the horizontal displacement of braced wall. The stability of the existing surcharged tunnel could be controlled by pre-loading on the braced wall. In this paper, it was investigated, if it would be possible to keep the existing surcharged tunnel stable by preventing the horizontal displacement of a braced wall by imposing the pre-loading during the ground excavation. For this purpose, large scale model tests were performed in a scale 1/10 at the test pit which was 2.0m in width and 6.0m in height and 4.0m in length. Isotropic test ground was constructed homogeneously by wet sand. Model tunnel was constructed in the test ground. Surcharge was loaded on the ground surface above the tunnel. During the tests, the behavior of model tunnel and model braced wall was measured. Numerical analyses were also performed in the same condition as the tests. And their results were compared to that of the model tests. Consequently, the effect of a surcharge could be compensated by imposing the pre-loading on the braced wall. The existing tunnel and the braced wall could be kept stable by preventing the horizontal displacement of the braced wall through pre-loading, although the tunnel is surcharged.

• Journal of Korean Tunnelling and Underground Space Association
• /
• v.9 no.4
• /
• pp.343-349
• /
• 2007

The reduced scale fire test provides basic data but it is not enough to analysis real fire problem directly because there is no exact analogy theory between a real fire and the reduced scale model. Therefore we have developed the large scale calorimeter in order to the real scale fire test. This advanced large scale calorimeter used for physical properties such as a heat release rate, based upon consumption of $O_2$ method. Using this large scale calorimeter, we cameo out the real scale vehicle fire test in order to evaluation for heat release rate. We obtained the calculated result for HRR $2.3{\sim}3.4\;MW$ and this result is very similar to the PIARC candidate HRR. It is approve that this facility has the reliability and it is capable of applying to the advance fire research in the future.

• Proceedings of the KSME Conference
• /
• 2007.05b
• /
• pp.2372-2377
• /
• 2007

Characteristics of turbulent lifted flames in coflow jets with the varying initial temperature have recently been investigated about only propane case diluted by nitrogen. The investigation has firstly improved a premixed flame model and a large scale mixing model among competing theories on the stabilization mechanism of turbulent flame to be suitable for a high temperature condition. In this research, about methane with good availability to apply for a practical combustor as clean fuel, its characteristics of turbulent nonpremixed flame have been studied experimentally. The results have shown an effectiveness of the premixed flame model and the large scale mixing model considered initial temperature variation. Additionally, considering the axial distance where the mean fuel concentration falls below the stoichiometric level along the center line of the jet according to diluting nitrogen, the premixed flame model have more accurately been improved.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2009.05a
• /
• pp.540-544
• /
• 2009

Hydrologic pattern under climate change has been paid attention to as one of the most important issues in hydrologic science group. Rainfall and runoff is a key element in the Earth's hydrological cycle, and associated with many different aspects such as water supply, flood prevention and river restoration. In this regard, a main objective of this study is to evaluate design flood using simulation techniques which can consider a full spectrum of uncertainty. Here we utilize a weather state based stochastic multivariate model as conditional probability model for simulating the rainfall field. A major premise of this study is that large scale climatic patterns are a major driver of such persistent year to year changes in rainfall probabilities. Uncertainty analysis in estimating design flood is inevitably needed to examine reliability for the estimated results. With regard to this point, this study applies a Bayesian Markov Chain Monte Carlo scheme to the NWS-PC rainfall-runoff model that has been widely used, and a case study is performed in Soyang Dam watershed in Korea. A comprehensive discussion on design flood under climate change is provided.

• Journal of The Korean Astronomical Society
• /
• v.29 no.spc1
• /
• pp.23-24
• /
• 1996

We reexamined CDM texture large-scale structure (LSS) formation model. We confirmed that texture model is consistent with 4-year COBE data both in an open and a critical matter density (${\Omega}_0$ = 1) universes, and then obtained normalization for density perturbation power spectrum. We next compare the power spectrum with LSS observation data. Contrary to the previous literature, we found that texture model matches with these data in an open universe no better than in an ${\Omega}_0$ = 1 universe. We also found that the model is more likely to fit these data in a cosmological constant dominated ($\Lambda-$) universe.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.48 no.4
• /
• pp.33-40
• /
• 2006

In present study, the principal stress condition needed to conduct cubical large-scale triaxial test which can reflect three dimensional stress condition (or plain strain condition) in a dam was investigated by performing model test and numerical analysis and the principal stress ratio varying with the height of CFRD was examined. Also, the principal stress ratio in CFRD body was investigated from the monitoring results of horizontal and vertical earth pressure gages, installed in the center zone and lower part of transition zone of the dam body, respectively, in order to consider the principal stress condition in the large-scale triaxial test to model the behavior of CFRD. The result of the study indicated that the principal stress ratio decreased gradually from the lower to the upper part in the dam body for its center axis and was about 0.5 and 0.2 in the lower and upper part, respectively.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.32 no.12
• /
• pp.970-976
• /
• 2008

Characteristics of methane turbulent non-premixed flame have been studied experimentally in coflow jets with initial temperature variation. The results showed that the premixed flame model and the large-scale mixing model for turbulent flame stabilization were effective for methane fuel considered initial temperature variation. Especially, the premixed flame model has been improved by considering nitrogen dilution for the liftoff height of turbulent lifted flame. In estimating blowout velocity and the liftoff height at blowout with the premixed flame model and the large-scale mixing model, the two turbulent models were excellently correlated by considering the effect of physical properties and buoyancy for the initial temperature variation.