[DECtalk] New synthesizer Gnuspeech

Sat Oct 24 04:27:41 EDT 2015

Re: throat temperature, It seem to have a very slight impact. I haven’t looked too much into it.
I don’t have problems understanding GNUspeech. Then again, I don’t have too much problems with the braille n’ speak synth either while a lot of people I talk to said it sounded very terrible and the h’s sounded like s’s on it, so maybe I’m just used to less clear synthesisers. There’s some things that I really like on this. One is intonation, which by the way is actually randomised. Very few synthesisers do this. So if you give it one short sentence, it will sound slightly differently every time. As far as pronunciation, the number preprocessing is probably the best I’ve seen, and goes up to… somewhere around 64 digits which are read as a regintillion. Prior to this the best I’ve seen was the Toshiba tts which could read half that.

> On 23 Oct 2015, at 22:40, Alex H. via Dectalk <dectalk at bluegrasspals.com> wrote:
> 
> Not missing the point mate, but such things are usually beat to death
> on this list so figured I'd chime in with my usual negativity. LOL :)
> 
> As for the pronunciation itself, it's spot on. I get excited for
> people to have more royalty-free speech so I get crabby when nobody
> does anything on the project or the quality is too poor to warrant
> reviving it. This by all accounts could end up being really solid and
> more pleasant than espeak and have a wider variety of applications
> like the medical and audiobook fields if it can keep up that rep of
> pronouncing things properly.
> 
> Changing timbres and the like shouldn't be too hard, as ideally one
> only need be able to grasp the rules and syntax, then tweak various
> stresses and so on to get it even better. Granted, what I prefer and
> what the guy across the street thinks is perfect spoken word timbre
> probably isn't going to be the same thing. Someone smarter than I
> would know about changing the actual frications of letters and
> clearing the beard from that guy's mouth hahaha.
> 
> Adding additional languages will be harder.
> 
> Piotir, does throat temperature do anything noticeable to the voice
> quality? That's an odd parameter. It makes me really think they've
> gotten the modeling philosophy down and don't mind throwing the curve
> ball in once in a while.
> 
> But yes for early days, I can say this already beats NVSpeechPlayer.
> It's impressive. That footprint is tiny too. 2 megs is nothing.
> 
> Alex
> 
> On 10/23/15, Tony Baechler via Dectalk <dectalk at bluegrasspals.com> wrote:
>> If I may say so, I think most of you are missing the point of the
>> synthesizer.  Yes, I'm the first to admit that I don't prefer the speech
>> quality and the inflection is a bit odd, but I noticed something which no
>> one else has commented on yet.  If you listen to the demo I created:
>> 
>> http://classicradio.us/iso/gnuspeech_gpl.mp3
>> 
>> you'll notice that it doesn't mispronounce anything.  I did not listen to
>> all of it and I noticed that it spelled every word in section 15, but I did
>> 
>> listen to several minutes of it and not once did I hear a word
>> mispronounced.  I think that's a particularly good piece of sample text
>> because it uses unusual phrases and words not found in every day English.
>> As much as I like DECtalk, IBM Viavoice and ESpeak, they all mispronounce
>> things.  The high end commercial synths seem to pronounce things correctly
>> most of the time, but they are big and really don't sound good at high
>> speed.  Granted, there doesn't seem to be a way to change the speech rate,
>> but considering the very small size (2 MB compiled if memory serves) and the
>> 
>> excellent pronounciation, it definitely has its points.  I'm specifically
>> thinking of small systems, like DTB players.  I tried the Rockbox speech and
>> 
>> didn't much like it, but I think Gnuspeech could be ideal for low memory
>> systems.  It took maybe a minute or two to convert the entire GPL to .wav.
>> Considering it's still early days, I'm impressed.
>> 
>> On 10/23/2015 1:21 AM, Alex H. via Dectalk wrote:
>>>> I hope this can be made into an NVDA addon please, I really like its
>>>> quality for some reason.
>>> 
>>> I don't at this stage, it sounds like Keynote Gold mixed with orpheus
>>> and a lot of missing teeth and drunken inflection.
>>> 
>>> Carlos, any idea when NVSpeechPlayer can sound like a true English
>>> speaker? I hate seeing forks and small projects get "just good enough"
>>> and then left to rot even though they're open source and nobody wants
>>> to pick up the ball (or like me can't code worth a cat's whisker).
>>> 
>>> If Gnuspeech were properly developed, it would be another option free
>>> option, and depending on i18/languages, could be really powerful for
>>> those looking for a TTS that speaks something other than English.
>>> Alas, dreams and fantasies probably.
>>> 
>>> Have a good weekend!
>>> Alex
>>> _______________________________________________
>>> Dectalk mailing list
>>> Dectalk at bluegrasspals.com
>>> http://bluegrasspals.com/mailman/listinfo/dectalk
>>> 
>> 
>> --
>> --------------------
>> Tony Baechler, founder, Baechler Access Technology Services
>> Putting accessibility at the forefront of technology
>> mailto:bats at batsupport.com
>> Phone: 1-619-746-8310   Fax: 1-619-449-9898
>> _______________________________________________
>> Dectalk mailing list
>> Dectalk at bluegrasspals.com
>> http://bluegrasspals.com/mailman/listinfo/dectalk
>> 
> _______________________________________________
> Dectalk mailing list
> Dectalk at bluegrasspals.com
> http://bluegrasspals.com/mailman/listinfo/dectalk