Noisy speech enhancement based on correlation canceling/log-MMSE hybrid method
In this paper, a speech enhancement method based on correlation canceling approach associated with the Log- minimum mean-square-error estimator is presented. Unlike the conventional statistical-model methods based on the nonlinear estimation of the enhanced speech signal, such as Maximum-Likelihood...
Gespeichert in:
Veröffentlicht in: | Multimedia tools and applications 2023-02, Vol.82 (4), p.5803-5821 |
---|---|
Hauptverfasser: | , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | In this paper, a speech enhancement method based on correlation canceling approach associated with the Log- minimum mean-square-error estimator is presented. Unlike the conventional statistical-model methods based on the nonlinear estimation of the enhanced speech signal, such as Maximum-Likelihood estimator (ML), Maximum A Posteriori (MAP) estimator, Minimum Mean Square Error (MMSE) estimator and log MMSE estimator, in the proposed hybrid method (CC/Log-MMSE), the nonlinear estimation is transformed into a linear estimation by exploiting the orthogonal projection of clean signal into the noisy signal. Thus, the enhanced signal represents the best “copy,” or estimate, of clean signal that can be made on the basis of the noisy signal vector. This is also seen as a canceling of the component of the noisy vector residing in the noise subspace, which therefore leads to improve the intelligibility of the enhanced signal. Extensive simulations are carried out using speech files corrupted by different noises available in the NOIZEUS corpus, show that the proposed hybrid method CC/Log-MMSE consistently outperforms the baseline methods of speech enhancement at different levels of SNR in terms of objective and subjective measures, spectrogram analysis and the overall SNR improvement. |
---|---|
ISSN: | 1380-7501 1573-7721 |
DOI: | 10.1007/s11042-022-13591-8 |