18 Apr
2020
18 Apr
'20
12:59 a.m.
Hi,
we are experimenting with doing remote voice tracking for Rivendell.
I have a test setup that is as close to my planned deployment setup as I
can get right now:
A Rivendell box running centos 7 and rivendell 3.2.1
A Raspberry Pi4 running Raspbian GNU/Linux 10
They are on the same switch very near each other and are connected together
audio wise like this:
Riv box onboard sound card
Pi4 usb UCA 222 wired up inputs to outputs
The Riv box is configured to use the alsa driver as the input and output
for rdlogedit.
The Pi is running jack using the dummy driver and zita bridges for the uca
222. (IIRC, I tried direct use of the USB codec and had more issues)
Jack Mixer is running to do routing and amplification as needed.
I could not find a voip phone that would work with jack so I am using
linphone with the pulseaudio-module-jack.
Using ssh tunnels with port forwarding to get toe the desktops of the Pi
and the Riv box.
After making all of the needed connections in the jack graph on the pi 4
and setting the linphone on the pi 4 to auto answer, we have a working
"system"
vnc to the Riv box (and the Pi if you like or want to tune volumes) and
bring up rdlogedit. Choose a lof that needs voice tracking and click the
button.
Dial the number/account for the liniphone on the pi4.
Start voice tracking. it works. (Mostly good enough. Latency wise. You can
slide things around to adjust. It is not bad.) it is the quality of the
vocal that out "ears" guy is not fully satisfied with. It will do in a
pinch but we would like to achieve better if we can get close to the same
latency with openob.
Moved on to try and replace the linphone link with a 2 way openob link over
openvpn. Latency was much worse and sound was "breaking up"...
I may have had problems with how I brought up the links.
The Pi4 may have been causing issues.
My network may have been less than perfect... ~;-)
Decided to move the testing location to the real location which is on
business fiber and not a home cable modem. Nook takes the place of the PI4.
Riv server is there but the nook and hte riv box are not wired up to allow
for the full test. I think I can fake enough on the nook to get a decent
test though.
On the far end, (same location) instead of a linphone call from a mac, we
have jack and openob on another nook running openvpn.
First the nook next to the Riv Box:
obuser@obstl01:~$ ps ax | grep openob
10773 pts/15 SLl+ 23:09 /usr/bin/python3 /usr/local/bin/openob 10.8.0.1
ntest-rx-node rem2shk-link rx -a jack -jn openobshackrx -aj -jp
jack_mixer:frmShk
10784 pts/7 SLl+ 35:25 /usr/bin/python3 /usr/local/bin/openob 10.8.0.1
stest-tx-node shk2rem-link tx -a jack -jn openobshacktx -aj -aj -jp
jack_mixer: 10.8.0.4 --framesize 2 -j 10 --fec -b128
Now the nook at the "remote" site:
obuser@obstl02:~$ ps ax | grep openob
8769 pts/2 SLl+ 62:32 /usr/bin/python3 /usr/local/bin/openob 10.8.0.1
ntest-rx-node shk2rem-link rx -a jack -jn openobnookrx -aj -jp
jack_mixer:frmShk
10740 pts/6 SLl+ 43:10 /usr/bin/python3 /usr/local/bin/openob 10.8.0.1
ntest-tx-node rem2shk-link tx -a jack -jn openobnooktx -aj -jp
jack_mixer:toShk 10.8.0.1 -e opus -b128 --framesize 2 -j 10 --fec
Could not get good latency. Help. I have reduced my frames per period on
both sides to see if that helps but have not been able to retest yet. Does
anyone see any glaring problems in my openob invocations at each end?
Should I reduce the quality of the Music link as that does not matter so
much for this application and leave the quality of the vocal link higher as
that is the point of trying this versus linphone?
Any other thoughts?
Does openob keep statistics on itself and its links in the redis server and
if so, can we get reports of the situation?
all the best,
drew
--
Enjoy the *Paradise Island Cam* playing
*Bahamian Or Nuttin* - https://www.paradiseislandcam.com/