Abstract

In this paper, we attempt to use pitch contour in Mandarin isolated words recognition by matching candidate pitch contour with a set of templates. To apply it in speaker-independent speech recognition, we propose a remedy algorithm to weaken speaker information in pitch contour. Speaker-dependent and - independent speech recognition experiments are carried out to evaluate performance of pitch contour and MFCC. Score fusion of MFCC and revised pitch contour are implemented. Experimental results demonstrate that pitch contour performs comparably with MFCC in speaker-dependent speech recognition. Through score fusion, the revised pitch contour helps in improving performance in speaker-independent speech recognition.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call