[asterisk-dev] iLBC packet loss concealment

[Steve Underwood]

Matthew Fredrickson wrote:

>
> On Apr 16, 2006, at 1:06 PM, Mike Taht wrote:
>
>>
>>
>> On 4/16/06, Steve Underwood <steveu at coppice.org > wrote:
>>
>>> at rates much lower than 64kbps.
>>>
>>> The main quality issue with normal phone calls is they are limited to
>>> 4kHz bandwidth. This is insufficient for good quality speech. 8kHz
>>> bandwidth really improves things. It lets you distinguish things "f"
>>> from "s", which is almost impossible on a normal phone line. In the
>>
>> This is one reason why good quality, in particular, continuous, voice 
>> recogition is nearly impossible over POTS. Even with limited domains, 
>> at 8khz it's all billions of dollars of R&D can do just to 
>> consistently mis-recognize "T-birds Pizza" in Los Gatos as "P-Birds 
>> Pizza".
>>
>> I note that this article on the wiki debunks the ilibc assumption for 
>> skype and goes into some depth:
>>
>> http://www.voip-info.org/wiki/view/Wideband+VoIP
>>
>> I went looking at mattf's asterisk wideband branch - seems out of 
>> date, is it dead?
>>  http://svn.digium.com/view/asterisk/team/mattf/asterisk-wideband/
>>
>
> Not entirely out of date.  I have a lot of work to do still and 
> haven't had much time to do it, but I think in a week or two I should 
> be able to get things moving again with it.  I just put in an 
> (untested) g.722 codec implementation and I'm working on a new 
> resampling codec at the moment.

I think that implementation of G.722 will take a lot of work to become 
useful. I played a bit with that when I implemented the version in 
spandsp. I could disclaim the one in spandsp, which passes all the ITU 
test vectors. The G.726 in spandsp also passes all the test vectors, and 
I think is the only free implementation with supports the 16kbps mode 
(though that sounds so bad, its isn't spectacularly useful). This 
includes handling a-law/u-law transcoding correctly.

Steve