17 May
2018
17 May
'18
2:04 a.m.
Le 16/05/2018 à 22:45, Chris Caudle a écrit :
The use-case is about voice calls. Humans are rather tolerant about the
latency of a phone call. isn't it?
I would set to the same sample rate that asterisk
expects so that no
rate conversion is needed.
Well, in fact Asterisk does not expect a specific sample rate. This
depends of the codecs that have been negotiated with the peer for the
ongoing call...
By default, the sip trunk uses ilbc g729 gsm g723 ulaw codecs. The
sampling rate can be 8kHz, 16kHz, or 13.3...
Maybe I can tell jack to use 16kHz sampling rate, and let it convert
when needed.
Linux xaal-c
4.15.17-1-MANJARO #1 SMP PREEMPT Thu Apr 12 17:29:48 UTC
2018 x86_64 GNU/Linux
That has the low-latency configuration set (PREEMPT), but
not the full
realtime patch (that would show PREEMPT-RT). Should be fine for most use,
but you may not be able to set to the very lowest latency settings. The
dummy backend uses 1024 sample period size by default. What app do you have connected?
The app is the speech-to-text Pocketsphinx (the code is slightly
modified for my needs)
Pocketsphinx uses pulseaudio as audio interface. So I need its jack
source/sink plugin.
So, there is no physical audio device in my system: the audio of the
caller is to be passed to the STT recognition system via the audio server.
Best regards
Christophe
17 May
17 May
8:11 a.m.
New subject: [Jack-Devel] How to use jackd as a 'system-wide' server?
On Thu, May 17, 2018 2:03 am, Christophe Lohr wrote:
OK, that is a little convoluted with all the different matching interfaces
required, but is actually a good use case for jackd and the dummy backend.
--
Chris Caudle
Maybe I can tell jack to use 16kHz sampling rate, and
let it convert
when needed.
Jack does no sample rate conversion, all ports always run at the sample
rate specified when jackd was started, but the comments in the asterisk
module indicated that asterisk would convert rates if needed. I did not
check to see if it was using one of the resampling libraries commonly used
on linux, but I would assume that audio quality and constant latency is
much less of an issue for telephony, so changing over to a more modern
resampler with higher quality e.g. zita-resampler would not really be
worth the effort.
The use-case is about voice calls. Humans are rather
tolerant about the
latency of a phone call. isn't it?
I would say yes up to a point, but 1024 sample buffer is very far under
that point. As long as you get no over or under runs with the PREEMPT
kernel you probably will not have a need to use a full RT patch.
What app do
you have connected?
The app is the speech-to-text Pocketsphinx (the code is slightly
modified for my needs)
Pocketsphinx uses pulseaudio as audio interface. So I need its jack
source/sink plugin.
2169
days inactive
2169
days old
jackaudio@lists.linuxaudio.org
1 comments
2 participants
participants (2)
-
Chris Caudle -
Christophe Lohr