Status of Pulse Audio in Speech Dispatcher

[Hynek Hanke]

Gilles Casse wrote:
>
>>  [...] in Pulse Audio and avoidance of having to reopen
>>    the PA connection on every synthesis request
>>     
>The patch which does not reopen the audio stream when a sound icon 
>follows speech at a distinct speech rate, this patch has not been 
>committed to cvs, because it is not yet compliant with pulseaudio.


I made changes to the PulseAudio backend by the 12th
of May (ChangeLog) and I thought this solves the problems,
at least partially. I do not know though whether there are
no side effects.

>>    - Pulse Audio support is still in an experimental stage and might
>>      not work with new versions of PA.
>> As I understand it, the pulse module adds latency compared to alsa but 
>> might be usable with pulseaudio.
>>     
This thing is a bit more complicated. The big added latency
(not speaking about reopening now) is not a necesary thing.
It is greatly due to the model that Pulse Audio was
using for communicating with the soundcard. I think there
was at least a 100ms delay.

This has all been solved in the new glitch-free version
of PA, but this is not yet released nor stable with SD
(it fails on failed assertions etc.).

You can read a bit about it here:

http://0pointer.de/blog/projects/pulse-glitch-free.html

Also, the current implementation hangs when switching
between text console and X-server when these two
environments are served by diffrent Pulse Audio servers
(as is necessary) due to some race conditions inside PA.

We are currently unable to get any further advice from
the PA developers and I think the most reasonable
thing now is just to wait patiently with our questions
and problems until PA 0.9.11 (glitch-free) is stabilized
and released.

So this is why I consider the PA backend experimental now,
although I think very little changes will be necessary after
everything is stabilized in PA 0.9.11.

With regards & thanks for testing,
Hynek