Hi, I am looking for some good method to estimate the fomants F1, F2. The method currently I am following is by Schafer and Rabiner, 1970. I am wondering how about the performance of such approach? And is there somewhere I can download the source code in C or Matlab? thanks a lot. Tracy ===== Laughter is an instant vacation. --Milton Berle |
|
Formant Estimation
Started by ●July 1, 2003
Reply by ●July 8, 20032003-07-08
hi Liu i don't know which is the method that you have chosen from Rabiner & Schafer, but anyway, here is what i think... take a frame of input speech. use the Levinson Durbin recursive algorithm to compute the alpha coefficients (coefficients of the short term predictor ie STP). This algorithm is given in Rabiner & Schafer as well as in the book by A. Kondoz. now you have the transfer function of the STP. The inverse of this is the transfer function of the short term "synthesis" filter. hence the frequency response of this filter is same as the envelope of the "spectrum" of speech under consideration. (we know that formants are defined as the peaks in the speech spectrum's "envelope") so if you can plot the freq response of the short term synthesis filter using MATLAB, your problem is solved. the first two peaks in this response are your formants f1 and f2. This way you can verify the results of your implementation. however, if you are already using the L-Durbin algo, i guess the results are fine anyway ;-) all you need is a C or Matlab code for Levinson Durbin algorithm, which, i think would be available on the internet. there is one more way to cross check the result using "Cool Edit" or similar tools : just select the appropriate section of the speech waveform, then click on "Frequecy analysis". This will show you the spectrum from 0 to 4 KHz. the first two peaks in its envelope are the required formants. hope that helps. best regards, Sameer. -----Original Message----- From: Chuping Liu [mailto:] Sent: Tuesday, July 01, 2003 11:53 PM To: Subject: [speechcoding] Formant Estimation Hi, I am looking for some good method to estimate the fomants F1, F2. The method currently I am following is by Schafer and Rabiner, 1970. I am wondering how about the performance of such approach? And is there somewhere I can download the source code in C or Matlab? thanks a lot. Tracy ===== Laughter is an instant vacation. --Milton Berle |