IMO you should open an issue even if it is not critical for your usage. This would allow this issue to be tracked. And since it seems to be a regression on the last alpha, it would allow core dev to fix it before 2020.1 release. Of course, core dev will not break again character reading (at least not intentionnaly).
De : email@example.com <firstname.lastname@example.org> De la part de Kostadin Kolev
In NVDA's settings, in the "Speech" category, there is an option called "Use spelling functionality if supported". You all probably know what this option is supposed to do, so I'm not going to bother explaining it.
Before NVDA 2019.3, this option was not working correctly for the most used bulgarian SAPI5 speech synthesizer "SpeechLab" (the voice "Gergana"), so we were disabling it on a new NVDA installation when using that voice. After the speech refactor in NVDA 2019.3 however, the option started working correctly for our voice. And by "correctly" I mean that with it disabled, separately read consonant characters from our cyrillic alphabet were spoken by "SpeechLab" as de vocalized (hope I'm using the most correct term here - it is hard for me to explain it), which is not the default behavior. With the option enabled, the consonants are read normally when read separately. And I think that is how the option should work.
In recent alpha snapshots of NVDA however, this option got broken again for voice "SpeechLab". No matter if the option is enabled or disabled, the synthesizer reads separate consonants as if the option is enabled. I think the option got broken after fixing the issue which prevented NVDA from reading single characters when the Speech Viewer was enabled.
I'm not filing a bug for now, since the problem is not critical and most users prefer "SpeechLab" to read consonants as if the option "Use spelling functionality if supported" is enabled, which NVDA does like so now anyway. But if there is a way to fix this without breaking character reading again, it would be nice.
Thanks much in advance!