`espeak-ng` or `speech-dispatcher` appends `-grandma` to default voice on Fedora 34

I am using a QGroundControl , QT based app, and have just moved to Fedora 34. On the new system, the voice sounded odd, like a grandpa speaking to me. I digged a little and made sure that the QT app sets locale correctly, and yes:

Then I digged a little more and this is what I have found in the log of speechd :

tail -f /run/user/1000/speech-dispatcher/log/speech-dispatcher.log | grep grand
[Tue Oct 26 21:22:59 2021 : 125478] speechd:     21:DATA:|SET SELF SYNTHESIS_VOICE English (America)+grandma
synthesis_voice=English (America)+grandma

After discussing this on the speech-dispatcher repo I am pretty sure that the problem is in my Fedora installation. The same .appimage works correctly on fresh Ubuntu 20.04. As a contributor of speechd wrote:

Maybe someone at fedora that packages either espeak or speech-dispatcher thought it would be humorous to have +grandma variant as the default or prioritize it when none have been selected explicitly.

How can I debug if further? The old versions of speech-dispatcher or espak-ng are not available for Fedora 34. I have also troubles with building it from source files.

1 Like

Hi,
I would suggest you get in touch with the package maintainer and let him/her know about the issue.

The fix seems easy and should be able to rebuild quite fast…
https://koji.fedoraproject.org/koji/packageinfo?packageID=6374

3 Likes

I’d be very surprised if this was the case. Anyway, I don’t see anything in the spec files that does this:

https://src.fedoraproject.org/rpms/espeak-ng/blob/rawhide/f/espeak-ng.spec

https://src.fedoraproject.org/rpms/speech-dispatcher/blob/rawhide/f/speech-dispatcher.spec

So, the first step would be to isolate if this is an issue with the package or some configuration issue. Can you please create a new user and see if the issue persists there? If yes, then it’s a bug that should be filed so that the maintainers can look into it.

I have just tried as a new user, same problem. Both configs look OK for me.
I am afraid that it’s a problem that the package maintainers will not be able to replicate. I need to build a simple app to test it.

I have just build a QT example app. This is the default voice it got from the speech-dispatcher:

@ankursinha I have just build the same example app on fresh Fedora live image and same happened. On Ubuntu live the Voice defaulted to normal English. This is definitely a Fedora issue.

I will fill the bug.

1 Like

That’ll be the best thing to do, yes. Thanks.

Please do share the link to the bug, and in the bug, please do provide detailed instructions on how to reproduce the issue. Otherwise it’ll be hard for the maintainers to debug.

I am really sorry: this is not a problem with Fedora. Fedora uses newer versions of speechd and espeak-ng than any other popular distro. The same problem will occur on newer distros. All Qt apps that does not specify the Voice will default to grandma version.

I will continue talks with developers of mentioned software packages and try to convince them that they should default to “normal” voice :slight_smile:

2 Likes