Speech Communication - Eurospeech '91
Speech Coding and Synthesis
Automatic silence/unvoiced/voiced classification of speech using a modified Teager energy feature
CONTROL'05 Proceedings of the 2005 WSEAS international conference on Dynamical systems and control
| Hi-index | 0.00 |
For wideband speech signal, if traditional noise cancelation algorithms are adjusted, it can be operated wrong way since unvoiced speech is similar as white Gaussian noise and high frequency on voiced speech that is over than MVF is also similar as noise in the presence of noise circumstance. So when speech signal is damaged by these reasons, it cannot be recognized exactly because that each phoneme is harmed and high frequency on voice speech signal is reduced even though the additive noise which is mixed with speech signal. It causes that speech signal makes like filtered by LPF. Therefore it is difficult to recognize since utterance is brushed. In this paper, we propose the noise estimation method and the detection of silence region for wideband speech signal in presence of noise by LPC analysis and spectral ratios. We compared speech signal which is mixed with additive white Gaussian noise with LPC prediction error signal, and when the value that the sum of speech signal is divided by the sum of LPC prediction error for each frame is closed with 1, we decided that the frame is noise-like speech region. And for this frame, if spectral ratios were upper than threshold, we decided that the frame is silence region. Finally, on this frame, we estimated noise level.