| |
 |
|
|
Science Forum Index » Compression Forum » multipulse based speech coding
Page 1 of 1
|
| Author |
Message |
| Rajesh Dachiraju |
 Posted: Thu May 01, 2008 8:39 pm |
|
|
|
Guest
|
hi,
I am experimenting on multipulse based speech coding, and am
facing a serious bug in my matlab code.
The energy(of 10 ms segments) of the modelled excitation is very
high compared to the actual LP residual. Is this
normal ? But after synthesis the energies are roughly same or if
not surprisingly the enegy of the original speech is
more than the reconstructed speech from modelled excitation.
I know there is serious bug in my code or if not i am missing out
something important.
Do we have to do any normalization of the residual and LPC
coefficients after computing them?
Also i have poltted the segmental SNR for different number of
pulses used and found that it is only 12 db
at 30 pulses per 5 ms. There is obviously something wrong with
this result.I am not using any perceptual
weighting and longterm prediction but still i feel the snr computed
is very low.
regards
Rajesh Dachiraju |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Wed Aug 20, 2008 5:53 am
|
|