抄録
The use of emotion-initiated gestures in human speech communication results in the improvement of speech understanding. However, this is a source of difficulty for automatic speech recognizers. In this paper, using the orderly changes found in the second formant, due to stress, a warping function is introduced that can be applied to the mel frequency scale during the calculation of MFCC parameters. We show that this approach leads to improvements in the stressed speech recognition results. Furthermore, using the second formant frequency as an extra element of the feature vector leads to further improvements in the speech recognizer performance.
