科研之友
微信
新浪微博
Facebook
分享链接
摘要
Warped Discrete Fourier Transform (WDFT) is able to increase frequency resolution at any selected parts of spectral axis without extending the number of sampling points. However, in ordinary WDFT synthesis block, inverse WDFT is imperfectly approximated since the WDFT related matrices are ill-conditioned. In this paper, we propose an innovative WDFT algorithm based on matching pursuit where an overcomplete set consisting of a series of complex exponent basis signals is constructed in order to minimize reconstruction error. The process is performed solely in the warped spectrum domain. The proposed algorithm is evaluated with speech signal reconstruction and used in whispered speech enhancement system. The experimental results show that signal can be perfectly synthesized without matrix inversion and the warped spectrum extracted using the proposed WDFT algorithm outperforms DFT and conventional WDFT based spectrum in whispered speech intelligibility enhancement.
