Speex dev - Dec 2010 - Problems with Speex Resamplers

How much latency are we talking about? It seems that this issue cannot be
easily pinpointed, but if it turns out to be related to the Speex Resampler
let me know and I will put it on my task list.

- Sherief

-----Original Message-----
From: speex-dev-bounces at xiph.org [mailto:speex-dev-bounces at xiph.org] On
Behalf Of Colin Guthrie
Sent: Monday, December 06, 2010 9:47 AM
To: speex-dev at xiph.org
Subject: Re: [Speex-dev] Problems with Speex Resamplers

'Twas brillig, and Sherief N. Farouk at 06/12/10 14:29 did gyre and
gimble:
> If you can produce a simple minimal repro that exhibits this issue on some
> procedural signal (sine wave, etc) I'd love to look into it some more -
I'm
> maintaining a hardened version of the Speex Resampler and would be very
> concerned if this bug made it into production.
Well at present, I'm not sure it is a bug at all.

I suspect that it is in some capacity related to PulseAudio's rewinding
that causes this. e.g. a new sounds starts that triggers a rewind, and
the resampler is reset and some other rewind happens and the latency of
the resampler during this reset phase is somehow longer than we need it
to be or something along those lines. After the rewind is dealt with,
then the latency of the resampler is not a problem any more... (and not
resetting and doing other such things also prevents overhead).

I'm not sure about the above hand-wavey, vague explanation but I suspect
it's something in that general area that's causing the problems rather
than a bug in speex per se.

Col.

-- 

Colin Guthrie
gmane(at)colin.guthr.ie
http://colin.guthr.ie/

Day Job:
  Tribalogic Limited [http://www.tribalogic.net/]
Open Source:
  Mageia Contributor [http://www.mageia.org/]
  PulseAudio Hacker [http://www.pulseaudio.org/]
  Trac Hacker [http://trac.edgewall.org/]

_______________________________________________
Speex-dev mailing list
Speex-dev at xiph.org
http://lists.xiph.org/mailman/listinfo/speex-dev

'Twas brillig, and Sherief N. Farouk at 06/12/10 14:52 did gyre and
gimble:
> How much latency are we talking about? It seems that this issue cannot be
> easily pinpointed, but if it turns out to be related to the Speex Resampler
> let me know and I will put it on my task list.
Will do.

Other resamplers in PA are not affected, but folk are looking into what
could be related issues just now so I'll be sure to report back on findings.

Col

-- 

Colin Guthrie
gmane(at)colin.guthr.ie
http://colin.guthr.ie/

Day Job:
  Tribalogic Limited [http://www.tribalogic.net/]
Open Source:
  Mageia Contributor [http://www.mageia.org/]
  PulseAudio Hacker [http://www.pulseaudio.org/]
  Trac Hacker [http://trac.edgewall.org/]

'Twas brillig, and Colin Guthrie at 06/12/10 15:10 did gyre and
gimble:
> 'Twas brillig, and Sherief N. Farouk at 06/12/10 14:52 did gyre and
gimble:
>> How much latency are we talking about? It seems that this issue cannot
be
>> easily pinpointed, but if it turns out to be related to the Speex
Resampler
>> let me know and I will put it on my task list.
> 
> Will do.
> 
> Other resamplers in PA are not affected, but folk are looking into what
> could be related issues just now so I'll be sure to report back on
findings.
Actually, at present it appears to have been triggered by a bug in the
gstreamer pulse output (which was my original thought, but I was
convinced this was an incorrect assumption)

It's strange that it causes problems on the speex resmaplers but not the
sinc or trivial resamplers, but at least there is a fix for now :)

I'll let you know if this fix is committed upstream.

Cheers

Col

-- 

Colin Guthrie
gmane(at)colin.guthr.ie
http://colin.guthr.ie/

Day Job:
  Tribalogic Limited [http://www.tribalogic.net/]
Open Source:
  Mageia Contributor [http://www.mageia.org/]
  PulseAudio Hacker [http://www.pulseaudio.org/]
  Trac Hacker [http://trac.edgewall.org/]