The last few days I have been experimenting with
using the Burg algorithm to estimate the coefficients
for my lowrate vocoder. The one I started many
The results I get using Durbin are much better than
with Burg so I will stick with Durbin for the moment.
I managed to fix some of the instability problems, they
were in the LSF quantisation routines. Now I need to
sort out the pitch tracking errors. I am thinking of using
analysis by synthesis to determine when I have made an
error in the pitch tracking/ voiced unvoiced decisions.
The vocoder is currently running at 2250 bps but I want
to reduce that further. I really need to use a split VQ
code book but building the code book is something I don't
think I have the facilities to do.
Still I am pleased with how far I have got so far but
development is very slow as I am no speech coding
expert! Still if it is not difficult it is not worth doing.