• Journal of Internet Computing and Services
    ISSN 2287 - 1136 (Online) / ISSN 1598 - 0170 (Print)
    https://jics.or.kr/
Menu

MCycleGAN: CycleGAN-based deep learning model for child speech extraction from noisy speech


Su-rak Son, Kyu-jeong Sim, Yi-na Jeong, Journal of Internet Computing and Services, Vol. 24, No. 3, pp. 1-8, Jun. 2023
10.7472/jksii.2023.24.3.1, Full Text:  HTML
Keywords: cycleGAN, multiple Cycle, Baby condition analysis, noise processing, spectrogram

Abstract

The most important technology for a robot to take care of a baby is to understand the baby's condition. Since babies mainly express their status through the pattern of crying sounds, research to classify the baby's status through voice is being actively conducted. Most of the studies that classify the baby's condition identified the baby's condition with a clean baby voice without noise. However, the baby's voice data collected in the real environment is likely to contain noise inside. Therefore, it is necessary to process the noise in the voice data. This paper proposes MCycle GAN (Multiple Cycle Generative Adversarial Net), which is a cycle GAN-based deep learning model for noise processing. MCycle GAN is a model in which multiple cycles are arranged in the existing Cycle GAN for more precise noise processing. The discrimination performance of the discriminator is improved by learning the adversarial relationship between a large number of generators and a small number of discriminators, and the generator needs to generate more precise forged data to deceive the discriminator. As a result of the experiment, the MCycle GAN model takes more training time than Cycle, but it showed stronger discriminant discrimination performance and generator forged data generation performance. However, when there are too many cycles, a small performance improvement can be seen compared to the increased learning time.


Statistics
Show / Hide Statistics

Statistics (Past 3 Years)
Multiple requests among the same browser session are counted as one view.
If you mouse over a chart, the values of data points will be shown.


Cite this article
[APA Style]
Son, S., Sim, K., & Jeong, Y. (2023). MCycleGAN: CycleGAN-based deep learning model for child speech extraction from noisy speech. Journal of Internet Computing and Services, 24(3), 1-8. DOI: 10.7472/jksii.2023.24.3.1.

[IEEE Style]
S. Son, K. Sim, Y. Jeong, "MCycleGAN: CycleGAN-based deep learning model for child speech extraction from noisy speech," Journal of Internet Computing and Services, vol. 24, no. 3, pp. 1-8, 2023. DOI: 10.7472/jksii.2023.24.3.1.

[ACM Style]
Su-rak Son, Kyu-jeong Sim, and Yi-na Jeong. 2023. MCycleGAN: CycleGAN-based deep learning model for child speech extraction from noisy speech. Journal of Internet Computing and Services, 24, 3, (2023), 1-8. DOI: 10.7472/jksii.2023.24.3.1.