Computer Science · Signal Processing
Acoustic noise suppression for speech enhancement
David V. Anderson, P. Whitehead, Mark A. Clements
See on the mapAbstract
Removal of noise from a single channel audio signal is becoming an increasingly important problem due to the proliferation of portable communication devices. Early methods of single channel noise suppression introduced artifacts that could be as annoying as the original noise. Substantial effort has been focused on more effective noise suppression with fewer artifacts, but many difficulties remain and the problems seem far from being solved. For example, nearly all noise suppression techniques actually reduce the intelligibility of the processed signal. This paper investigates the sources of some of the artifacts common with Wiener filtering and proposes a structure to mitigate or eliminate these artifacts and effectively reduce noise without adversely affecting intelligibility. Results of one implementation of such a structure demonstrate significant improvements both in speech recognition performance and perceived speech quality.