Abstract | ||
---|---|---|
The major drawback of the most widely used spectral subtraction (SS) algorithm is, it fails to reduce musical noise. In addition to this in SS, the subtraction rules are mainly based on false assumptions about cross-terms are being zero. A novel approach is proposed to overcome these shortcomings in the SS algorithm. A technique is implemented to calculate exactly the cross-terms which involve the differences in phase amidst degraded speech signal and noise model. The proposed technique gain function is having the same properties as traditional minimum mean square error (MMSE) algorithms. The experimental results on NOIZEUS speech corpora reveal that the proposed algorithm outperforms the traditional SS algorithms in terms of speech quality and intelligibility at lower SNR conditions. Further, the output of the proposed approach shows that there is no audibility of musical noise in processed or enhanced speech sound. The numerical complexity computation and pictorial representation of input-output waveforms and corresponding spectrograms of proposed and existing speech enhancement techniques are also presented in this work. |
Year | DOI | Venue |
---|---|---|
2022 | 10.1007/s11042-022-12152-3 | Multimedia Tools and Applications |
Keywords | DocType | Volume |
Spectral subtraction (SS), Noisy speech data, Enhanced speech data, Musical noise suppression, Cross-terms, Spatial procedure (SP) | Journal | 81 |
Issue | ISSN | Citations |
17 | 1380-7501 | 0 |
PageRank | References | Authors |
0.34 | 9 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Thimmaraja G. Yadava | 1 | 0 | 0.34 |
B. G. Nagaraja | 2 | 0 | 0.34 |
H. S. Jayanna | 3 | 0 | 0.34 |