The independent component analysis (ICA)-based speech features were used to remove Gaussian noise in the noisy speech signals using a maximum a posteriori (MAP) estimator. Speech signals corrupted by additive white Gaussian noise were recovered with greatly improved signal-to-noise ratio (SNR) values. Denoised spectral features also resulted in better recognition rates than the standard MFCC features.
ASJC Scopus subject areas
- Electrical and Electronic Engineering