Re: I take it there have been no improvements to the flawed espeak ng in 2017.4?
|
Brandon Cross <bcross3286@...>
I don't know where to begin, so I'll just pick a point in all this thread and go from there. It is not logical to assume the developers will make a change for just one person, therefore, might I suggest creating an issue on github with specific problems bullet pointed, and ask everyone in the list to look at this and reply to the issue if they have similar problems. To whoever was saying whether or not Eloquence dictionaries will be applied to eSPeak ones or not, no, definitely not, Eloquence has hard and fast rules and is most inflexible in changing its pronunciation, eSpeak bends more easily, so the same modifications you would need to make eSpeak say a name correctly would most likely not work in Eloquence. I would recommend against buying a sapi addon unless it will only be used for long reading, as navigating the system and so forth, these voices have too much latency to be practical. I would also recommend against purchasing the vocalizer / eloquence addon for NVDA made by code factory. This addon is quite frankly junk. The Eloquence portion of it crackles and pops, and this behavior has been tested on two machines. It also uses extremely long pauses at punctuation marks like a comma or period. It cannot be configured to be shorter. One option left to the OP is to obtain NVDA from source code via github and they would then have the ability to place whatever version of eSpeak they would wish to use, then build it. This is a long view scenario, and would require certain knowledge of how to clone a git repository, as well as using scons to build it. Also, the repo doesn't come with scons, so finding that is another challenge. I don't use eSpeak, I don't like it, can't stand it, never could. Now, aside from that, I have noticed an unpleasant quality of the voice that appears in most variants since the switch to NG. I cannot describe this quality accurately. Suffice it to say that it appears almost as a random series of very light clicks is injected into the speech stream. This is highly distracting as it just sounds terrible. Again, this has been tested on at least 3 machines and found to be the case among all of them. Keeping in mind that this machine has no trouble running higher quality human sounding concatinative synths, so I doubt its a performance issue. I have cooked up my own variants and they are affected as well. Maybe I will record it, because I can't put it into words, it is definitely not the voice crackling, or breaking up, it is as if someone hijacked the audiostream, modifying it memory to include these tiny little clicks and then let it pass to the sound card that way.
On Tue, Mar 27, 2018 at 10:43 AM, Ervin, Glenn <glenn.ervin@...> wrote: If I run it too fast, I may understand it all, but I cannot retain the information, my puny brain needs time to process it for retention.
|
|
|