Error with OneCore voice and automatic language switching in NVDA2022.1

[CyrilleB79]

Context

Issue found while trying to reproduce #13727. Opened as a new issue as per #13727 (comment).
Cc @seanbudd

Steps to reproduce:

• Use OneCore non-English voice (mine is Hortense - French)
• Ensure NVDA has automatic language switching enabled (during my tests, automatic dialect switching is disabled however)
• Open the following URL: data:text/html,<p>Hello</p><p lang="en">world</p>
• Read the page

Actual behavior:

• With NVDA 2021.3.4 (portable), the whole page is read in French since I have no English OneCore voice.
• With NVDA 2022.1 (installed), the word "world" is not read and an error is logged.

Here are the logs for both versions of NVDA:

• OneCoreWithNVDA2021.3.4.log: OK
• OneCoreWithNVDA2022.1.log: Error

Expected behavior:

OneCore reads the whole page; if no English voice is available the default (French) voice should be used instead.

!additional note

If reading 5 times the line "world" causing the error, the synth switches to eSpeak. Cf log:
OneCoreSwitchToESpeak.log

System configuration

NVDA installed/portable/running from source:

Installed

NVDA version:

2022.1

Windows version:

Windows 10 2004 (x64) build 19041.388

Name and version of other software in use when reproducing the issue:

N/A

Other information about your system:

• Windows French, NVDA language default (thus French)
• In the list of OneCore voices, I have French voices (Hortense, Julie, Paul) and German voices (Stefan, Katja, Hedda) but no English voice. I do not remember if I had never had one, if I have uninstalled it (what is very unlikely but why not for tests), or if this voice has disappeared without my willing (e.g. during an update or an installation of something).

Other questions

Does the issue still occur after restarting your computer?

Yes

Have you tried any other versions of NVDA? If so, please report their behaviors.

As written earlier, 2021.3.4 (portable) does not exhibit this issue.

If NVDA add-ons are disabled, is your problem still occurring?

Test done without add-ons

Does the issue still occur after you run the COM Registration Fixing Tool in NVDA's tools menu?

Yes

[seanbudd]

Hi @CyrilleB79 - I've tried reproducing this with a fresh French Windows 10 installation and had no success. I was using a slightly newer build as your build is not available. Would you be able to updating Windows? Can you try installing and removing English voice packs? Do you know if anyone else has experienced this issue?

[seanbudd]

Is this reproducible with 2022.1beta1?

[seanbudd]

I think this aspect of the logs shows that OneCore is partially installed:

DEBUGWARNING - synthDrivers.oneCore.OneCoreSynthDriver._isVoiceValid (15:50:50.999) - MainThread (27140):
Could not open registry key HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech_OneCore\Voices\Tokens\MSTTS_V110_enUS_DavidM, FileNotFoundError(2, 'Le fichier spécifié est introuvable', None, 2, None)
DEBUGWARNING - synthDrivers.oneCore.OneCoreSynthDriver._isVoiceValid (15:50:51.000) - MainThread (27140):
Could not open registry key HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech_OneCore\Voices\Tokens\MSTTS_V110_enUS_ZiraM, FileNotFoundError(2, 'Le fichier spécifié est introuvable', None, 2, None)
DEBUGWARNING - synthDrivers.oneCore.OneCoreSynthDriver._isVoiceValid (15:50:51.000) - MainThread (27140):
Could not open registry key HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Speech_OneCore\Voices\Tokens\MSTTS_V110_enUS_MarkM, FileNotFoundError(2, 'Le fichier spécifié est introuvable', None, 2, None)

[CyrilleB79]

I will try to reproduce the issue with NVDA 2022.1beta1

My Windows build is unavailable since I should actually update it. For some reason I have an error and cannot update; I have to take some time to investigate why.

I have had no report of someone else experiencing this issue. I have probably had issue while installing or uninstalling things since the log indicates traces of David, Mark and Zira English US voices.

If you really want I can try to install/uninstall these voices again. But after that, I may be unable to reproduce the issue again.

Actually, my concern is not that I have English voices missing (since I have not yet tested to reinstall them).

My concern regards the following points:

1. There is a regression: my configuration causes an error and no speech with 2022.1 whereas in 2021.3.x the whole text is read without any error.
2. These tests seems to indicate that with NVDA 2021.3.x, NVDA was able to fallback to current voice when auto-switch tried to switch to an invalid voice; with NVDA 2022.1 this causes an error instead.

Additional note: I have not tested, but maybe removing the registry key of a voice on a good install can help test this code path.

[seanbudd]

Hi @CyrilleB79 - thanks for these notes. I was planning to try and find a way to reproduce this by doing something similar to what you suggested.

I agree that this is still a concerning regression, but if it is not widely spread it has a lower impact than initially considered.

[CyrilleB79]

@seanbudd, I have just tested with NVDA 2022.1beta1 (run from launcher). The issue was already present in this version.

Also, I agree with you: since I am the only one known so far to be impacted by this issue and since I have not yet tested to reinstall and uninstall these voices, the issue is not critical at all.
Moreover, there is a work-around: disable automatic language switching.

[seanbudd]

@CyrilleB79 - re-opening this as we would still like to fix this