![]() |
ITRW on Non-Linear Speech Processing
|
![]() |
This paper presents a wavelet-based technique to increase the quality and naturalness of LPC based synthesized speech signals. The proposed method is based on wavelet decomposition. We first obtain the wavelet coefficients, and then the variances of the wavelet coefficient at the last four scales (correspond the higher frequency region) of the synthetic speech are replaced by the original variances of the original speech. We apply the technique to synthetic speech. The results suggest that the wavelet-based technique increases the naturalness of the synthesized speech.
Bibliographic reference. Atas, Mehmet / Baykut, Süleyman / Akgül, Tayfun (2007): "A wavelet-based technique towards a more natural sounding synthesized speech", In NOLISP-2007, 11-14.