Hello,
I conducted an experiment in MATLAB related to noise suppression.
I have a noisy speech segment X=S+N where S is clean speech and N is
noise.
I did LPC analysis of a frame of X and obtained an estimate of a noisy
excitation (residual) signal u_noisy.
I also did LPC analysis of a frame of S and obtained an estimate of a
clean excitation signal u_clean.
I then filtered u_noisy with the LPC-coefficients from the LPC-analysis
of the clean speech frame.
I also filtered u_clean with the LPC-coefficients from the LPC-analysis
of the noisy speech frame.
Result:
The experiment showed that having clean LPC coefficients doesn't really
make a big difference. What makes a big difference is if you have a clean
residual signal.
So I am wondering if my results are correct? Or did I oversee something?