Speech enhancement using critical band spectral subtraction

This paper proposes a new enhancement technique for the enhancement of broadband noise corrupted speech. The technique exploits the human auditory systems inability to distinguish between individual frequency components within critical frequency bands. Spectral subtraction is used and the spectrum is considered as critical frequency bands rather than individual frequency components. The proposed technique is compared with the existing spectral subtraction technique, using both subjective and objective speech assessment measures. Results are quoted and indicate that there is a significant increase in intelligibility and quality.