Re: [orca-list] Speech-dispatcher/orca integration specification, first draft.

[Hynek Hanke <hanke brailcom org>]

Jacob Schmude napsal(a):
> Unix domain sockets wouldn't fix this, all this would do is eliminate
> the requirement for sd to communicate over TCP locally, meaning it would
> be a bit faster

I don't think it would be faster. This is a really small difference
and not even important for the speech synthesis stuff. Orders of 
magnitude more time is being wasted in other places. Pulse audio had
a huge ,,latency'' last time I checked due to some problems
with kernel configuration (described in PA mailing lists),
IPC in general (regardless of mechanism) seems to be much slower
than expected for transfer of larger data in most current Linux 
distributions, I guess it is the same problem as described on
tha PA mailing list.

Even in communication with Festival, we could save about
half of the total time for ,,synthesis'', if the transfer
of data back to Speechd (implemented as pipes) was ,,fast''.

See

[pulseaudio-discuss] Distribution kernels and glitch-free

> and would not go down if the networking subsystem dies.

A lot of things go down in that case. I wouldn't worry about
this that much.

> As far as I know, Pulseaudio would be the only way to handle the
> per-client audio streams as things stand now, though in theory support
> for doing this via ALSA alone could be coded into sd directly.

I believe Pulse Audio is the best way for the future.

With regards,
Hynek Hanke