Modified Speech Separation Deep Learning Network Based on Hamming window

Speech separation is attracting widespread interest due to the sound mixing in real environments in and out door applications. Although the researchers have used many algorithms, the separation rate in the real environment is still poor. This paper presents speech separation using a modified Deep learning neural (DLN) algorithm. Interestingly, the modification has reduced the complexity of the original DLN algorithm, while, high value of separation rate has been gained caused by using Hamming instead of Hanning windows against the other algorithms. The separation rate reaches 98.6%, while, the advancement over the nearest algorithm is 2.8%.