speech parcor
HI all
I have a conclusion problem in my final term project (speech processing).
In this project I want to add normal noise {n(0,.01)} to LPC, PARCOR and LSF coefficients and check the theory that says PARCOR and LSF coefficeins are more roboust in copare with LPC coefficients.
I have 5 sound files (phonems) including a vowel, nasal, fricative, affricative and stop which are windowed by a hamming window ( length of window is 256 with 50% overlap). then I compute the LPC coefficents in a matrix with N rows and M coloums ( each coloumm contains the LPC coef for one
windowed frame), then I used LPC to PACOR and LPC to LSF converter functions to get these coeffs. the I added same noral noise to LPC, PARCOR and LSF coeffs for L=4 , L=14 and L=50. I plotted the results and I put then In the attached file. (for example noisyLPC=.1*noise*LPC+LPC and ...)
The fft of windowed frame is plotted in green , noisy coefficeints in red and noise-free coeffsare plotted in blue.
The main question is why LSF coeffs are so weak after adding noise?
and why PARCOR and LSF coeffs are not roboust in noise???