Well, I don't quite know how eSpeak works, but ETI Eloquence does an amazing job when it comes to correctly pronouncing compound words.
As for phonemes, I don't have older versions of eSpeak, but I may try NVDA's older versions and hear how the previous versions of eSpeak sound.

If I somehow happen to be able to learn to edit these phoneme files, then I will do my best to make eSpeak speak realistically in as many languages as I can.

Well, the creation of the phonemes themselves is really hard, but there are so many by now, you may find fitting ones among the exsisting ones. When it comes to defining phonemes to be used for the rules, that's rather easy once you get the hang of it. I unfortunately haven't got the time to outline all this here now, but from what I get, there's ESpeak's documentation that you can look at for this.

As for Eloquence, I think it uses a different system for text processing than Espeak. I don't know if espeak may be able to convert to this, but from what I get, Espeak can only accept one to one rules, and there's no way to tell it boundaries between words or something. Adding every single word possible to the rules would be obviously rather tedious, unless you develop a script which does the job.

Espeak has a lot of problems with the Russian language. It is a fact. But I'm not sure if a not native speaker should write about such problems. I am a philologist. If you want, you can write me a message and we will discuss how best to send suggestions for improving phonetics.

Do we have the eSpeakNG documentation embedded in NVDA or anywhere else? Where to find it?
--Cytat (hozosch):
Well, the creation of the phonemes themselves is really hard, but there are so many by now, you may find fitting ones among the exsisting ones. When it comes to defining phonemes to be used for the rules, that's rather easy once you get the hang of it. I unfortunately haven't got the time to outline all this here now, but from what I get, there's ESpeak's documentation that you can look at for this.

--Koniec cytatu

May we not refer to an already existing and comprehensive dictionary that has all the words used in German, or any other language that we wish to work on?
--Cytat (hozosch):
As for Eloquence, I think it uses a different system for text processing than Espeak. I don't know if espeak may be able to convert to this, but from what I get, Espeak can only accept one to one rules, and there's no way to tell it boundaries between words or something. Adding every single word possible to the rules would be obviously rather tedious, unlesss you develop a script which does the job.

--Koniec cytatu

All I know is that the Russian eSpeak does not pronounce number 1 correctly.
--Cytat (monstricek):
Espeak has a lot of problems with the Russian language. It is a fact. But I'm not sure if a not native speaker should write about such problems. I am a philologist. If you want, you can write me a message and we will discuss how best to send suggestions for improving phonetics.

--Koniec cytatu

By the way, speaking of eSpeak, I'd appreciate your feedback on my voices for it. Posted them in the My Apps thread.

Monstricek, even a not native speaker hears if something's pronounced wrong. Let me say an example. I'm learning russian for almost 3 months now, and I'm already able to hear if vocalizer says something wrong.

So how do we use the Mbrola voices with the new eSpeak NG that comes installed along with NVDA 2019.3 and later?
Why I'm specifying the version is that I couldn't find a Python 3 version of the Mbrola add-on.

Hmmm, I wonder if you just take the espeak.dll file from there, and create a mbrola in espeak.py file based on the latest espeak.py file, which should be in NVDA's source code.

Well, great...
But I have no Python knowledge, and I guess I have to ask whether it's just copy and paste or it requires any coding?
By the way, what I actually meant was to directly embed Mbrola into eSpeak NG. I mean, using the Mbrola voices as eSpeak NG voices.
--Cytat (hozosch):
Hmmm, I wonder if you just take the espeak.dll file from there, and create a mbrola in espeak.py file based on the latest espeak.py file, which should be in NVDA's source code.

--Koniec cytatu

Oh, well, there, I'm as unexperienced as you. But I don't like how they sound in ESpeak anyway, so that's not a thing for me.

I'm not an ESpeak fan at all, but I did notice, after downloading the latest ESpeak NG, that I only had one English voice and it said weird stuff after everything it said, so I uninstalled it. It said something like bookmark mark followed by an ever-increasing number. It was a British voice at that, not an American one.

Yes, the sapi5 version has really obvious issues. There are really some basic problems going on, some of these are also not at all easy to fix, at least that's what I'm inclined to believe when I look at this:
https://github.com/espeak-ng/espeak-ng/issues/7

Oh wait, I just checked the issue again, and it seems that Reece has made a to do list of things with checkboxes only he can check or uncheck. I can see that this isn't complete yet.

Can we download e-speak ng for sapi and use it now?

You can, but it isn't as good, and I use the word loosely, as what's included in NVDA.

Hi guys, i need help. I want to use Siri voices for my NVDA, because i love the german voice Martin and the american english voice Aaron. Does someone have the link?