Optimization of the Speaker Recognition in Noisy Environments Using a Stochastic Gradient Descent

Keywords:
Speech recognition, deep neural network, stochastic gradient descent, noise environments.

Abstract:
Noise-robust speech recognition system is still one of the ongoing, challenging problems, since these systems usually work in the noisy environments, such as offices, vehicles, airplanes, and others. Even though deep learning algorithms provide higher performances, there is still a large recognition drop in the task of speaker recognition in noisy environments. The proposed system is tested on VIDTIMIT dataset in the presence of Additive White Gaussian Noise changing the Signal-to-Noise Ratio levels. The experimental results show how the optimization of Stochastic Gradient Descent algorithm parameters such as learning rate and dropout rate, can improve the performance of speech recognition in both noisy and less noisy environments.

CITATION:

IEEE format

Copy to clipboard

A. Nasef, M. Marjanović Jakovljević, A. Njeguš, “Optimization of the Speaker Recognition in Noisy Environments Using a Stochastic Gradient Descent,” in Sinteza 2017 - International Scientific Conference on Information Technology and Data Related Research, Belgrade, Singidunum University, Serbia, 2017, pp. 369-373. doi:10.15308/Sinteza-2017-369-373

APA format

Copy to clipboard

Nasef, A., Marjanović Jakovljević, M., Njeguš, A. (2017). Optimization of the Speaker Recognition in Noisy Environments Using a Stochastic Gradient Descent. Paper presented at Sinteza 2017 - International Scientific Conference on Information Technology and Data Related Research. doi:10.15308/Sinteza-2017-369-373

BibTeX format
Download

RefWorks Tagged format
Download