Alsa/Pulseaudio conflicts

[Gilles Casse]

Gilles Casse wrote:
> Hynek Hanke wrote:
>> The current implementation of Pulse Audio in Speech Dispatcher
>> is slow because the implementation of it contains a problem.
>> For each sound output, it opens and closes the whole connection,
>> so there is a big overhead which is completely unnecessary.
> 
> [...]
> 
>  From pulseaudio 0.9.8, pa_stream_update_sample_rate has been added for 
> updating the rate. But if the format of the sound icon is different from 
> the speech samples, I guess that the stream must be still closed and 
> reopened.
> 
> I have updated the code consequently but unfortunately can't go further 
>   at the moment. A patch is copied below as is (but code is not yet 
> operational).
> 

The patch might work but is not yet compatible with pulseaudio 0.9.10 .
It will be tested against 0.9.11 or possibly using the development 
version of pulseaudio.
More details below.

Best regards,
Gilles

* pulseaudio 0.9.10:

This release prevents the use of PA_STREAM_VARIABLE_RATE (in pulseaudio, 
src/pulse/stream.c, create_stream function, CHECK_VALIDITY, the 
PA_STREAM_VARIABLE_RATE flag is missing).

Flag present in the development version of pulseaudio (git).

* pulseaudio 0.9.10 + fix (test using speechd-el + espeak or ibmtts)

The switch from sound icon to speech is around 100ms faster, but after a 
few seconds an assert occurs in pulseaudio.

* pulseaudio from git

The sources need speexdsp > 1.2, not yet available on the tested system 
(Ubuntu Hardy).