Re: nuance cerence neural tts for NVDA?
Hi.toggle quoted messageShow quoted text
I have to admit that I'm not impressed by these voices. They're too jerky, even more so than the previous expressive versions. Even the conversational one is pretty bad about it, though i I like how it allows the artifacts of her voice to show through, you can actually hear her vocal artifacts like at one point her voice croaked a little, but it was clearly the real deal rather than concatenation problems. The built in google voices on Android are starting to get it right, the male one is expressive and reasonably smooth while some of the vocal artifacts are being represented.
In my opinion, most of the concatenated voices are not reproducing very well. They're nearly all jerky, and their intonation is just goofy. And that goes for everybody. About the only ones who don't screw up the inflections is the ones that don't actually do much of it to begin with, Cepstral comes to mind.
For Nuance, a couple of the later versions of Tom are pretty good, as is Samantha, but the rest meh.
Ivona is amazing at not being jerky, but they somehow are difficult to understand, like they're just muttering along or something although Sally isn't quite so bad as Joey and the others. I vaguely think the Alexa voice is based off of one of the Ivona voices but I don't know that for a fact. She is an incredible voice, but being a female, I understand very little that she says, significantly reducing the actual usefulness of the echo devices for me. I do much better using the male voice for the google home devices. I prefer the one called pink although it has been a while and that voice may not be called that anymore.
Out of all voices I still find Eloquence and ESpeak, in that order, to be the easiest to understand for my hearing. I even have difficulty understanding DeCtalk. The only DeCtalk voices I really understand well are the earliest ones like 1.8, and the speechmaster2000 that sounds vaguely similar to very old DeCtalk. Anything newer has a really odd artifact where it starts tumbling when it gets to read anything that is longer than 5 words or so. The syllables get shortened and it stops anunsiating clearly, which, by the way, is another thing that I dislike about the Nuance voices they all tend to do that too, one notable exception is Tom. Apple's Alex voice does this badly although it can be somewhat reduced by running at a very slow speed. It's like those voices get distracted with what they are reading and they forget to read out clearly, and so they just start muttering to themselves. Ivona's voices did not do that.
One of my favorite voices that I forgot to mention is the LH True Voice peter. Very plane, sharp, smooth and consistent the only problem was all the spelling and the excessive inflection when encountering exclamation marks. That is one somewhat human sounding voice that I can actually understand well enough to read emails with.
The main reason that I use Android is because I can put Eloquence on it. Using the iPhone with the Alex voice is doable, but tends to be a lot more stressful just for struggling to understand him. If I'm in a car there is an even wider gap between them, because I am so good at understanding Eloquence that I only need bits and pieces to make out what is being said where other voices I really need to hear everything very clearly and even that's no guarantee. Of course, ESpeak is also good enough to use I just hate the sound of it.
RH voice is getting there, it vaguely reminds me of the above mentioned LH True voice, but it's too much bass and too little mids and highs. They're a bit quiet. I don't find myself easily able to read emails with them yet.
Aaron Spears, AKA Valiant8086 General Partner at Valiant Galaxy Associates "we make (VERY GOOD AUDIOGAMES) for the blind comunity" http://valiantGalaxy.com
On 9/11/2020 1:14 PM, Gene wrote:
The web page where you can read about the voices and find samples is: