Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
SHO based Deep Residual network and hierarchical speech features for speech enhancement
Ist Teil von
International journal of speech technology, 2023-07, Vol.26 (2), p.355-370
Ort / Verlag
New York: Springer US
Erscheinungsjahr
2023
Link zum Volltext
Quelle
SpringerLink (Online service)
Beschreibungen/Notizen
The human frequently finds difficulty in understanding the speech due to the real-world noises. The presence of external noises corrupts the listening comfort of user. Hence there is a need for the enhancement of speech. In this paper, the Shepherd Harmony Optimization (SHO)-based Deep Residual network (DRN) is developed for speech enhancement. Here, the developed SHO-based DRN is the combination of the Shuffled Shepherd Optimization Algorithm (SSOA) and Harmony Search optimization (HS). The Hanning window is used for the pre-processing of the input data. In this method, the Bark Frequency Cepstral Coefficients (BFCC) and Fractional Delta amplitude modulation spectrogram (FD-AMS) are used for the feature extraction. Moreover, the noises present in speech signals are predicted for eliminating the distorted noises and external calamities. Besides, the DRN classifier is utilized to improve the speech signal. The classifier is trained by newly devised optimization algorithm. Besides, the developed speech enhancement technique obtained better performance in terms of Perceptual Evaluation of Speech Quality (PESQ) with 2.646 and Root Mean Square Error (RMSE) with 0.0067.