JCSE, vol. 20, no. 2, pp.70-77, 2026

DOI: http://dx.doi.org/10.5626/JCSE.2026.20.2.70

Research on English Speech Enhancement in Complex Noisy Environments through a Deep Learning Algorithm

Jingning Li
School of Basic Science, Yangzhou Polytechnic Institute, Yangzhou, Jiangsu 225127, China

Abstract: The quality of speech can be greatly affected in complex noisy environments. This paper proposes an algorithm that combines a gated dilated convolutional block (GDCB) and a recurrent convolutional neural network (RCNN), called GDCB-RCNN, for speech enhancement, and then verifies its enhancement effect using two English speech datasets, VoiceBank+DEMAND and TIMIT+NOISEX-92. The results showed that on the VBD dataset, the GDCB-RCNN algorithm had a perceptual evaluation of speech quality of 3.25, a short-time objective intelligibility of 0.96, and a complex speech intelligibility gain of 4.33, a compound background noise invasion degree of 3.66, and a compound overall speech quality of 3.79, respectively, outperforming some other methods. It also demonstrated excellent performance in various noisy environments. Moreover, it was found that accurate ratio masking outperformed the traditional ideal ratio masking. The results demonstrate the effectiveness of the GDCB-RCNN algorithm for English speech enhancement, which can be further applied in practice.

Keyword: Deep learning; Noisy environment; Speech enhancement; Convolutional neural network

Full Paper:   5 Downloads, 8 View