Summary: | 碩士 === 國立中央大學 === 電機工程研究所 === 83 === Basing on pitch period , we divided each recognized
utterance into three segments unvoice-segment , voice-
segment , nonvoice-segment ; and then each proper sub-
segment is searched to comput Linear Prediction Coding ( LPC
) ceptral. However , in white noise condition , the
recognition rate will be degraded sharply . In this thesis ,
there are two properties were utilized to deal with the white
noise speech : (1) Zero mean value of white noise . (2)
The voice-segment and nonvoice-segment have the property of
quasi-period . According to these two properties , we can
eliminate the white noise properly and the general
property of speech wave will still be retained .
|