After using spd-say (with pico TTS) with the root account, spd-say stops working for users

[Jean-Philippe Guérard]

The command definition (GenericExecuteSynth) in /etc/speech-dispatcher/modules/pico-generic.conf use a fixed file name for the wave file ($TMPDIR/pico.wav):

GenericExecuteSynth \
"pico2wave -w $TMPDIR/pico.wav -l $VOICE \'$DATA\' && $PLAY_COMMAND $TMPDIR/pico.wav

If you use spd-say or a speech dispatcher enabled application with root, it will create the pico.wav file owned by root with restricted permissions.

If you try to use spd-say with a different user, it will then fail, failing to access the /tmp/pico.wav file.

A solution would be to create a temporary file per user ID:

GenericExecuteSynth \
"USERID=`id -u` && umask 077 && pico2wave -w $TMPDIR/pico-$USERID.wav -l $VOICE \'$DATA\' && $PLAY_COMMAND $TMPDIR/pico-$USERID.wav"

[Luke Yelavich (themuso)]

Have you tested this fix? I only ask because I am not sure shell constructs like environment variable declaration work in that context.

[Jean-Philippe Guérard (fevrier)]

Yes, the fix is tested and works nicely.

[Luke Yelavich (themuso)]

I've actually got a better idea. Instead of checking the user id, we can use XDG_RUNTIME_DIR as a base for the wav file. So something like $XDG_RUNTIME_DIR/speech-dispatcher. That will be cleaned up when the user logs out, and nobody then knows that another user is using speech-dispatcher's generic synth mechanism. Probably not really a security issue, but is still cleaner as this file is temporary for only the logged in user, and only for their session.

I'll go ahead and implement this unless you have an objection.

[Jean-Philippe Guérard (fevrier)]

That seems like a good idea.

[Jean-Philippe Guérard (fevrier)]

I did some additional tests. This does not work in all cases.

If I switch to root with a "sudo su -", XDG_RUNTIME_DIR is not defined.