Hello,

[Tim Cross]

Samuel Thibault <samuel.thibault at ens-lyon.org> writes:

> Didier Spaier, le lun. 01 oct. 2018 21:20:26 +0200, a ecrit:
>> It seems complicated to check in module_utils_addvoice.c the
>> availability of each voice listed in the config file, because the paths
>> to the voice files are not the same for all synthesizers.
>
> This could be made configurable in the .conf file, e.g. something like
>
> VoiceFileDependency "/usr/share/mbrola/$VOICE/$VOICE"
> VoiceFileDependency "/usr/lib/espeak-data/voices/mb/mb-$VOICE"
>
> If no such line appears, all voices would be listed. If such line
> appears, the voice addition would be done only if all these files exist.
>
> for cases where the paths would be unsure, these can be left as comments
> and distributions can fix them according to their own correct paths.
>
> (things should really "just work" without having to modify configuration
> files, and automatically modifying configuration files seems fragile to
> me)
>

I agree that modifying configuration files is fragile and writing
scripts to do this as part of different package management systems
represents additional work which is likely to discourage maintainers.

My suggestion would be to have this sort of configuration broken out
into a speechd.conf.d directory. The main configuration logic would then just
'source' files in this directory as if they were parts of the main
configuration file. This is a common pattern package maintainers are
likely to be familiar with and it gives them the ability to add to the
configuration without having to modify any existing files - just drop a
new one in. It could be done so that anything in the .d directory
overrides whatever is in the main speechd.conf file, making that file a
generic default and providing an easy 'back to factory' reset i.e. just
remove files from the .d directory and your back to defaults.

Tim

-- 
Tim Cross