Speech Dispatcher and audio

[Hynek Hanke <hanke brailcom org>]

Luke Yelavich wrote:
> 2. The pulseaudio problems are fixable, it just requires a bit of work to re-adjust the buffering metrics used in the speech synthesizer drivers

Hello Luke and all,

we did such experiments last year, and we communicated with Lennart,
tested several patches, kernels etc. and though we got some improvements,
they couldn't compare to ALSA in speed or reliability. There are some
principal problems:

1) For this kind of architecture, where there are multiple
processes in the audio chain, we need low latency
and pre-emption in kernel as explained by Lennart
and also documented in Jack. Distributions don't normally
offer this to stay compatible with some proprietary drivers.

2) Pulse Audio itself (not the backend) needed some serious work
on analysing and fixing issues related to buffer metrics
and the glitch-free model at that time, but the main developers
were unwiling to devote significant time to the issues important
for accessibility.

I don't know if something changed in these two points from then,
but unless we are sure about it, it doesn't seem to be reasonable
to spend more time on PulseAudio.

Since the problem is in the architecture or PulseAudio itself,
I don't think experiments using Gstreamer on top of PA or using
Jack instead of PA and such can help it.

Now we tend to think that any solution where the audio
is not output to kernel space directly from the synthesis
driver (like we do with ALSA) is not a possible way given
the current state of things.

Of course you are welcome to cross check these results,
that would be very useful. There might be something which
changed or something that we are missing.

With regards,
Hynek Hanke