Better visualization for formant analysis by means of time-frequency distributions

摘要

This work explores the possibility of using time-frequency distributions (TFD) to extract time varying formant information. This technique makes use of the TFD of Cohen's (see Prentice-Hall, Englewood Cliffs, NJ, 1995) class and provides the profile of formant variation continuously in the time-frequency plane which can be employed to improve formant tracking and formant bandwidth estimation. The performance of this method is compared with other existing methods, which have their own pitfalls, using a modulated synthetic signal as input. It is shown the proposed method gives better formant estimation and also provides better visualization representation. This method is used to analyze real human speech and the results can be helpful for speech understanding and speech synthesis.