• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2018.10a
• /
• pp.369-372
• /
• 2018

This paper suggests non-parallel-voice-conversion network conversing voice between unmapped voice pair as source voice and target voice. Conventional voice conversion researches used learning methods that minimize spectrogram's distance error. Not only these researches have some problem that is lost spectrogram resolution by methods averaging pixels. But also have used parallel data that is hard to collect. This research uses PPGs that is input voice's phonetic data and a GAN learning method to generate more clear voices. To evaluate the suggested method, we conduct MOS test with GMM based Model. We found that the performance is improved compared to the conventional methods.

• Journal of Microbiology and Biotechnology
• /
• v.31 no.6
• /
• pp.890-901
• /
• 2021

The 11α-hydroxylation of canrenone can be catalyzed by Aspergillus ochraceus in bioreactors, where the geometry of the impeller greatly influences the biotransformation. In this study, the effects of the blade number and impeller diameter of a Rushton turbine on the 11α-hydroxylation of canrenone were considered. The results of fermentation experiments using a 50 mm four-blade impeller showed that 3.40% and 11.43% increases in the conversion ratio were achieved by increasing the blade number and impeller diameter, respectively. However, with an impeller diameter of 60 mm, the conversion ratio with a six-blade impeller was 14.42% lower than that with a four-blade impeller. Data from cold model experiments with a large-diameter six-blade impeller indicated that the serious leakage of inclusions and a 22.08% enzyme activity retention led to a low conversion ratio. Numerical simulations suggested that there was good gas distribution and high fluid flow velocity when the fluid was stirred by large-diameter impellers, resulting in a high dissolved oxygen content and good bulk circulation, which positively affected hyphal growth and metabolism. However, a large-diameter six-blade impeller created overly high shear compared to a large-diameter four-blade impeller, thereby decreasing the conversion ratio. The average shear rates of the former and latter cases were 43.25 s^-1 and 35.31 s^-1, respectively. We therefore concluded that appropriate shear should be applied in the 11α-hydroxylation of canrenone. Overall, this study provides basic data for the scaled-up production of 11α-hydroxycanrenone.

• Journal of the Korean Society of Combustion
• /
• v.19 no.4
• /
• pp.49-55
• /
• 2014

Design criterion for the size of micro Pt-catalytic combustor is investigated in terms of heat release rate. One-dimensional plug flow model is applied to determine the surface reaction constants using the experimental data at stoichiometric butane-air mixture. With these reaction constants, the mass fraction of butane and heat release rate predicted by the plug flow model are in good agreement with the experimental data at the combustor exit. The relationship between the size of micro catalytic combustor and mixture flowrate is introduced in the form of product of two terms-the effect of fuel conversion efficiency, and the effect of chemical reaction rate and mass transfer rate.

• Journal of Korea Multimedia Society
• /
• v.25 no.11
• /
• pp.1572-1581
• /
• 2022

Energy conversion to renewable energy is being promoted to solve the recently serious environmental pollution problem. Solar energy is one of the promising natural renewable energy sources. Compared to other energy sources, it is receiving great attention because it has less ecological impact and is sustainable. It is important to predict power generation at a future time in order to maximize the output of solar energy and ensure the stability and variability of power. In this paper, solar power generation data and sensor data were used. Using the PCC(Pearson Correlation Coefficient) analysis method, factors with a large correlation with power generation were derived and applied to the GAM(Generalized Additive Model). And the prediction accuracy of the power generation prediction model was judged. It aims to derive efficient solar power generation in the future and improve power generation performance.

• Journal of information and communication convergence engineering
• /
• v.21 no.1
• /
• pp.9-16
• /
• 2023

The natural language on social network platforms has a certain front-to-back dependency in structure, and the direct conversion of Chinese text into a vector makes the dimensionality very high, thereby resulting in the low accuracy of existing text classification methods. To this end, this study establishes a deep learning model that combines a big data ultra-deep convolutional neural network (UDCNN) and long short-term memory network (LSTM). The deep structure of UDCNN is used to extract the features of text vector classification. The LSTM stores historical information to extract the context dependency of long texts, and word embedding is introduced to convert the text into low-dimensional vectors. Experiments are conducted on the social network platforms Sogou corpus and the University HowNet Chinese corpus. The research results show that compared with CNN + rand, LSTM, and other models, the neural network deep learning hybrid model can effectively improve the accuracy of text classification.

• Nuclear Engineering and Technology
• /
• v.36 no.1
• /
• pp.1-11
• /
• 2004

Two approximate methods for a cosmic radiation shielding calculation in low earth orbits were developed and assessed. Those are a sectoring method and a chord-length distribution method. In order to simulate a change in cosmic radiation environments along the satellite mission trajectory, IGRF model and AP(E)-8 model were used. When the approximate methods were applied, the geometrical model of satellite structure was approximated as one-dimensional slabs, and a pre-calculated dose-depth conversion function was introduced to simplify the dose calculation process. Verification was performed with mission data of KITSAT-1 and the calculated results were also compared with detailed 3-dimensional calculation results using Monte Carlo calculation. Dose results from the approximate methods were conservatively higher than Monte Carlo results, but were lower than experimental data in total dose rate. Differences between calculation and experimental data seem to come from the AP-8 model, for which it is reported that fluxes of proton are underestimated. We confirmed that the developed approximate method can be applied to commercial satellite shielding calculations. It is also found that commercial products of semi-conductors can be damaged due to total ionizing dose under LEO radiation environment. An intensive shielding analysis should be taken into account when commercial devices are used.

• Journal of the Korean Institute of Telematics and Electronics D
• /
• v.36D no.2
• /
• pp.81-89
• /
• 1999

Among various wavelength conversion technologies, that using the cross-gain modulation in laser diode makes it possible to deal with the high speed signal quite simply and efficiently. In this paper, presented was the applicability of an improved time-domain large-signal dynamic model as a CAD tool to analyzed the characteristics of SSGDBR(Superstructure Grating Distributed Bragg Reflector) laser diodes used for wavelength converters. Using this model, it was shown that this kind of wavelength converter can provide the widely tunable wavelength conversion of the high speed data above 10 Gbps. We also investigated the effect of input optical power and the bias current on the characteristics of the device such as extinction ration and eye diagram. The modeling results show very similar trend to the experimental reports.

• Journal of Life Science
• /
• v.21 no.1
• /
• pp.127-133
• /
• 2011

${\beta}$-Galactosidase was immobilized on chitosan bead by covalent bonding using glutaraldehyde. The characteristics of the immobilized enzyme were investigated. Maximum immobilization yield of 75% was obtained on chitosan bead. Optimum pH and temperature for the immobilized enzyme was 7.0 and $50^{\circ}C$, respectively. The immobilized enzyme showed a broader range of pH and temperature compared to a free one. A mathematical model for the operation of the immobilized enzyme in a packed-bed reactor was established and solved numerically. Under different inlet lactose concentrations and feed flow rate conditions, lactose conversion was measured in a packed-bed reactor. The experimental results of continuous operation in a packed-bed reactor were compared to theoretic results using Michaelis-Menten kinetics with competitive product inhibition and external mass transfer resistance. The model predicted the experimental data with errors less than 5%. Process optimization of continuous operation in a packed-bed reactor was also conducted. In a recirculation packed-bed operation, conversion of lactose was 97% in 3 hours. In a continuous packed-bed operation, the effect of flow rate and initial lactose concentration was investigated. Increasing flow rates and initial lactose concentration decreased the conversion of substrate.

• Journal of Microbiology and Biotechnology
• /
• v.18 no.6
• /
• pp.1130-1135
• /
• 2008

The effect of pH on anaerobic hydrogen production was investigated under various pH conditions ranging from pH 3 to 10. When the modified Gompertz equation was applied to the statistical analysis of the experimental data, the hydrogen production potential and specific hydrogen production rate at pH 5 were 1,182 ml and 112.5 ml/g biomass-h, respectively. In this experiment, the maximum theoretical hydrogen conversion ratio was 22.56%. The Haldane equation model was used to find the optimum pH for hydrogen production and the maximum specific hydrogen production rate. The optimum pH predicted by this model is 5.5 and the maximum specific hydrogen production rate is 119.6 ml/g VSS-h. These data fit well with the experimented data($r^2=0.98$).

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.7
• /
• pp.303-314
• /
• 2023

Voice conversion, a technology that allows an individual's speech data to be regenerated with the acoustic properties(tone, cadence, gender) of another, has countless applications in education, communication, and entertainment. This paper proposes an approach based on the StarGAN-VC model that generates realistic-sounding speech without requiring parallel utterances. To overcome the constraints of the existing StarGAN-VC model that utilizes one-hot vectors of original and target speaker information, this paper extracts feature vectors of target speakers using a pre-trained version of Rawnet3. This results in a latent space where voice conversion can be performed without direct speaker-to-speaker mappings, enabling an any-to-any structure. In addition to the loss terms used in the original StarGAN-VC model, Wasserstein distance is used as a loss term to ensure that generated voice segments match the acoustic properties of the target voice. Two Time-Scale Update Rule (TTUR) is also used to facilitate stable training. Experimental results show that the proposed method outperforms previous methods, including the StarGAN-VC network on which it was based.