Research on Methods of Speech Separation Based on Deep Learning
CAO Weifan
ZHANG Erhua
Abstract：Aiming at the problem of monaural speech signal-to-noise separation,a method based on deep learning is used.The speech information contains in the high-frequency part and the low-frequency part of the speech is compared.Only the low-fre-quency part of the speech is used to train the deep learning model,and better separation results are obtained.According to the differ-ence of the separation effect of the speech with stationary noise or paroxsive noise,the phenomenon of crosstalk in the separation re-sults is studied.Griffin-lim algorithm is used to reconstruct the separated speech and smooth the amplitude to eliminate channeling.
Keywords：signal-noise separationdeep learningcrosstalk phenomenon
Publication Date：2025-10-20
Online Publishing Date：2026-01-16(First online date of this platform, not the publication date of the document)
Pages：4( 2693-2696 )