Publication Details
Noise Estimation for Spectral Subtraction in Speech Processing
noise estimation, noise subtraction, spectral subtraction, speech recognition
This contribution presents two often used algorithms for enhancement of noisy speech signal in spectral domain. These algorithms do not differ in removing noise from the input speech. Different parts are processes of noise estimation which are used before noise suppression. Both algorithms have ability to track non stationary noise signal and do not need voice activity detector. Both methods have been tested in front-end distributed speech recognition (DSR) system. Algorithm parameters, as CPU complexity or memory requirements, as well as the need of delay in processing, will be discussed.
Noise Estimation for Spectral Subtraction in Speech Processing.
@INPROCEEDINGS{FITPUB7060, author = "Petr Motl\'{i}\v{c}ek", title = "Noise Estimation for Spectral Subtraction in Speech Processing", pages = 4, booktitle = "Proceedings of 8th Conference STUDENT EEICT 2002", year = 2002, location = "Brno, CZ", publisher = "Faculty of Electrical Engineering and Communication BUT", ISBN = "80-214-2116-9", language = "english", url = "https://www.fit.vut.cz/research/publication/7060" }