12 Jun
2012
12 Jun
'12
3:45 a.m.
I believe I found the source of the remaining warble. The unvoiced synthesizer seems to
be filtering the same noise in each 20ms frame, rather than starting with fresh white
noise each time, resulting in a 50 Hz hum. I'll try fixing it to match the spec over
the few days.
- Clayton
--- In op25-dev(a)yahoogroups.com, "argilo314" <argilo@...> wrote:
I found the correct fix for the phase issue in the fine transition code. There was pi^2
in a couple places where it should have been 2*pi. Here's a patch that includes all
my fixes so far:
Index: blocks/src/lib/software_imbe_decoder.cc
===================================================================
--- blocks/src/lib/software_imbe_decoder.cc (revision 301)
+++ blocks/src/lib/software_imbe_decoder.cc (working copy)
@@ -757,6 +757,7 @@
{
int i,j;
//initialize
+ OldL = 0;
Old = 1; New = 0;
psi1 = 0.0;
for(i=0; i < 58; i++) {
@@ -931,7 +932,7 @@
// if(abs((int)sample) > 32767) {
// sample = 32767 * (sample < 0) ? -1 : 1; // * sgn(sample)
// }
- sample /= 16384.0; /* 32768.0 - but audio amplitude is not full range of int16
*/
+ sample /= 65536.0;
samples->push_back(sample);
}
}
@@ -1394,10 +1395,8 @@
if(vee[ell][ New]) {
if ( vee[ell][ Old]) {
if(ell < 8 && fabsf(w0 - Oldw0) < .1 * w0) { // (fine
transition)
- const double PI_SQUARED = M_PI * M_PI;
- const double INV_PI_SQUARED = 1.0 / PI_SQUARED;
Dpl = phi[ell][ New] - phi[ell][ Old] -(Oldw0 + w0) * ell * 80;
- Dwl = .00625 * (Dpl - PI_SQUARED * floorf((Dpl + M_PI) *
INV_PI_SQUARED));
+ Dwl = .00625 * (Dpl - 2 * M_PI * floorf((Dpl + M_PI) / (2 * M_PI)));
THa = (Oldw0 * (float)ell + Dwl);
THb = (w0 - Oldw0) * ell * .003125;
Mb = .00625 *(MNew - MOld);
--- In op25-dev(a)yahoogroups.com, "rrgsti" <bobrich@> wrote:
I admire your tenacity Clayton!
The change to software_imbe_decoder does improve audio quality at this end as well. I
added a small amount of logging and noted that the code is going back and forth between
the fine and coarse transition operations, so it doesn't seem like something is just
'pegged' at a wrong value. This is all magic to me, however, so I have no
intuition about what is or is not correct.
This is running on 11.10 with itpp 4.07 built from source, latest copies of op25 and
fairly recent gnuradio and gr-baz.
Thanks!
--- In op25-dev(a)yahoogroups.com, "argilo314" <argilo@> wrote:
I've made a bit of progress. It appears that the phase calculation in
software_imbe_decoder::synth_voiced is not correct in the "fine transition"
case, resulting in abrupt phase shifts between the 20ms frames. I'm not sure yet how
to correct it, but by commenting out the "fine transition" case completely so
that the "coarse transition" case always runs instead, I've eliminated most
of the warble.
Another problem I noticed is that the audio output is sometimes greater than 1.0,
resulting in clipping if connected directly to an audio sink. I corrected this by
changing the "sample /= 16384.0;" line at the end of
software_imbe_decoder::decode_audio back to "sample /= 32768.0;".
- Clayton
--- In op25-dev(a)yahoogroups.com, "argilo314" <argilo@> wrote:
I'm getting the same issue with the sound: speech is intelligible but with a strong
warble that makes it sound like talking through a fan. I had a quick look at the waveform
and spectrum of the output but didn't spot any obvious problems there. Not knowing
IMBE, I couldn't get far with the code. Any ideas?
- Clayton
--- In op25-dev(a)yahoogroups.com, "rrgsti" <bobrich@> wrote:
>
> Basically looking at bit 113 (only) seems to indicate whether or not it's a
voice frame of some sort. The audio still sounds like everyone is talking through a fan,
but there are no dropouts, no clicks, no stutters, etc.
>
> I could not find anything resembling a DUID or NAC in the headers. If anyone has a
suggestion to de-warble the audio, I would greatly appreciate it.