Speech recovery based on the linear canonical transform

As is well known, speech signal processing is one of the hottest signal processing directions. There are exist lots of speech signal models, such as speech sinusoidal model, straight speech model, AM–FM model, gaussian mixture model and so on. This paper investigates AM–FM speech model by the linear canonical transform (LCT). The LCT can be considered as a generalization of traditional Fourier transform and fractional Fourier transform, and proved to be one of the powerful tools for non-stationary signal processing. This has opened up the possibility of a new range of potentially promising and useful applications based on the LCT. Firstly, two novel recovery methods of speech based on the AM–FM model are presented in this paper: one depends on the LCT domain filtering; the other one is based on the chirp signal parameter estimation to restore the speech signal in LCT domain. Then, experiments results are presented to verify the performance of the proposed methods. Finally, the summarization and the conclusion of the paper is given.