loss of speech in speakup when switching between console and gui

[Samuel Thibault]

Hello,

(I reordered things a bit to make the story clearer for pulseaudio
maintainers in Cc)

Didier Spaier, le ven. 02 nov. 2018 01:13:09 +0100, a ecrit:
> This message is an answer to the thread started by:
> https://lists.debian.org/debian-accessibility/2018/10/msg00000.html
> 
> @Keith: If you don't use pulseaudio, the issue could be an unexpected
> consequence of applying since Thu, 03 May 2018 the patch audio-pause:
> "Pause espeak when the console is switched to a graphical VT"

Well, I believe that report is just another case of the well-known issue
that once pulseaudio is started in X (e.g. for orca), it holds the ALSA
card completely and espeakup can't take it again.  The pause patch makes
espeakup release the ALSA card so that pulseaudio triggered by Orca can
take it. This is considered a better behavior than not getting any audio
inside X just because espeakup holds the ALSA card.

> I then made these changes:
> 1) Edit /etc/pulse/default.pa to append these lines:
> load-module module-alsa-sink device=dmix
> load-module module-alsa-source device=dsnoop

So using dmix is not the default in Debian?

> 2) In /usr/share/alsa, remove the files pukse-alsa.conf and
> alsa.conf.d/alsa.conf, to avoid setting pulseaudio as the default plugin for
> applications using Alsa when pulseaudio is running.

> I made these changes so that applications using pulseaudio and applications
> using alsa directly can nicely coexist, not stepping on each others toes.

> I don't know if the modifications I made are acceptable by the Debian
> authorities though ;)

There is no such thing like "Debian authorities".
There are the maintainers of the pulseaudio stack, which define a
default configuration which aims at the most common case.  I don't know
why dmix is not part of it, that's with them to be discussed, e.g. in a
bug report. Making pulseaudio share the device with alsa thanks to dmix
seems like an option indeed, that you could document on
http://wiki.debian.org/accessibility
I don't know what counterparts there might be to it, again pulseaudio
maintainers will know better.

> I also disabled autostarting of pulseaudio at the user level, appending:
> "Hidden=true" to the file /etc/xdg/autostart/pukseaudio.desktop but maybe that
> doesn't matter. Anyway pulseaudio is spawned by the applications that need it. 

And notably here by Orca, so I don't think that is involved here.

> But these changes were not sufficient to solve the issue so I had a look at the
> speakup Debian package. Seeing the aforementioned patch I thought that it could
> cause the issue. To check I just replaced /usr/bin/espeakup by the binary
> shipped in Slint and it worked.

Ok, so somehow espeakup doesn't manage to take the ALSA card again once
pulseaudio is started in X?  It'd be interesting to check with the patch
(i.e. the Debian binary)

- whether starting espeakup only after running pulseaudio in X works (in
  which case it's the espeakup resume which fails).

- a backtrace of espeakup when it failed to resume, i.e. attach a gdb to
  it and run thread apply all bt full. One such kind of trace was posted
  on http://linux-speakup.org/pipermail/speakup/2018-October/061491.html
  I haven't found the time to really look at it yet, various things have
  kept popping up.

> I understand that you won't be interested by my settings of alsa and pulseaudio
> as you don't use pulseaudio, but this could also solve the issue mentioned in
> the thread "pulseaudio and espeakup" beginning with this message :
> https://lists.debian.org/debian-accessibility/2017/12/msg00089.html

Yes, thus documenting on the wiki, so people can configure it even if
pulseaudio maintainers prefer not to set it by default.

> Oh, and I almost forgot: with the patch when rebooting from Mate the system
> didn't halt but was stuck with this message (from systemd, I assume):
> As stop job is running for Software speech output for Speakup
> This do not happens anymore after having replaced the espeakup binary by the one
> shipped in Slint.

That's an interesting point indeed, it really sounds like the daemon is
getting stuck somehow.

Samuel