A FILTERING TECHNIQUE FOR BEST FREQUENCY ESTIMATION
A.Tali*, A.El Moussaoui
Université Abdelmalek Essaadi, Faculté des Sciences, Tétouan
* Corresponding author. E-mail :email@example.com
Received : 11 May 2001; revised version accepted 26 September 2001
In many speech applications, one often confronts the problem of extracting useful frequencies buried in a noisy signal. Linear Prediction Coding (LPC) method as applied to speech has been well used for spectrum determination, but does not always gives the exact position of the poles on the areas of interest, and it also gives rise to the predominance of lower frequencies preventing exact determination of the highest. To overcome these problems, we propose a new method combining the ESPRIT algorithm (Estimation of Signal Parameters via Rotational Invariant Technique) and the LPC method in its simple form, which will be referred to as a Combining Method. The first one is used to derive special filters in order to isolate only one dominant spectral component. The filtered spectrum is then processed by the LPC method to estimate the frequency of interest.
Application of this method to formant determination of voiced speech signals gives good results compared to direct LPC analysis.
Keywords : Filtering; Frequency estimation; Low-bit rate LPC.