[External] Re: Audio/voice conference

[Honckiewicz, Filip BIS]

Thanks for the response Nicolas, so I'll concentrate on Farstream understanding.

Cheers!
Filip

-----Original Message-----
From: gstreamer-devel [mailto:gstreamer-devel-bounces at lists.freedesktop.org] On Behalf Of Nicolas Dufresne
Sent: Wednesday, March 26, 2014 5:35 PM
To: Discussion of the development of and with GStreamer
Subject: [External] Re: Audio/voice conference

Le mercredi 26 mars 2014 à 10:46 +0000, Honckiewicz, Filip BIS a écrit :
> But the main problem is how to receive audio from many clients and, 
> secondly, how to mix the sound together? Should one client be a server 
> which takes many input sounds, mix it and broadcast? I suppose 
> broadcasting from every client would be problematic because of echo, 
> etc. and it's not a good idea. I also think that creating separate 
> pipeline for n-clients (n could be 20, 30) would kill devices (ARM 
> based), cause as I observed, one receive pipeline when using 
> gst-launch takes up to 10% of CPU (with PCMA codec). How implementing 
> this using gstreamer libraries would lower CPU usage comparing to my 
> gst-launch test pipelines?

The performance highly depends on your latency configuration. Though it's hard to give advise on optimization so early.

> 
> Or maybe stop to even think about using gstreamer directly and use 
> farstream like it was suggested by Jan Schmidt in "GStreamer for 
> massive real-time voice chat" discussion?
> 

I would recommend looking into that. Farstream already covers all of the use cases you seem to list. See the example code, and you can probably replicate all your test pipelines base on that.

cheers,
Nicolas