[DECtalk] voice creation

Thu Sep 7 15:53:14 EDT 2017

Yes, some of the Circum Reality voices sound good. It's also wonderful that
Jake Gross made a SAPI keynote. I did not know that I could record myself
saying a word for correct intonation. Is version of Circum Reality on Jake's
site the most recent version? My only issue with CR on Jake's site is lack
of uninstaller. Removing a installed CircumReality voice requires editing
the registry. I had to do that when I had CR voice(s) on my computer some
time ago. Editing the registry scares me because of the damage incorrectly
editing registry entries can do to a computer. Is there a newer version of
CR available which uninstalls a voice/ removes voice/CR registry keys?

Blake

  _____  

From: Dectalk [mailto:dectalk-bounces at bluegrasspals.com] On Behalf Of
Brandon Tyson
Sent: Thursday, September 07, 2017 12:47 PM
To: DECtalk
Subject: Re: [DECtalk] voice creation

Hello,

I wanted to touch on one point please:

Even if they can't do SAPI, and even though this wouldn't be the same, I'm
still quite happy with the now open-sourced CircumReality voice and think
that they sound pretty good.

What I particularly like about those is that even if they do have a
mispronunciation, it's giving you the possibility to change the way a
particular word, or even whole phrases, sound through manual hand made
recordings, I think, which would theoretically be used in conjunction with
the synthesizer itself.

Does this make sense?

E.g. A user types in "Wow!" but they find the intonation is incorrect. It
says "wow." Instead, making it sound boring. So the user can go in and
manually correct this to have it say "Wow!" correctly by making an
individual recording for that.

Thanks,

Brandon

From: Devin Prater <mailto:r.d.t.prater at gmail.com> 
Sent: Thursday, September 7, 2017 8:56 AM
To: DECtalk <mailto:dectalk at bluegrasspals.com> 
Subject: Re: [DECtalk] voice creation

Yuly, GPU is the Graphics processing unit, and the CPU is the central
processing unit. 

A neural net is is a way for computers to learn as a human does, through
artificial intelligence. 

Sent from my iPhone

On Sep 7, 2017, at 7:39 AM, Ulysses Harmony Garcia via Dectalk
<dectalk at bluegrasspals.com> wrote:

What is GPU, and how is that different from CPU? Also, how does neural
networking work? Does the computer map out a brain on a series of
coordinates the way it does when mapping out an MRI or 3D printing the model
of the brain? I'm interested in knowing what programming languages,
interpreter scripts, etc they use so that people in college and university
labs who do neuroscientific work can build more neural networks to
experiment on.
I've read some sci-fi literature on how mind-uploading could work, but the
problem is that it would be nearly impossible to calculate the exact number
of neurones, neurotransmitters, synapses, and receptors, and which memories
or part of identity they link up to. It would require a huge load of
processing and a really high knowledge of map coordination in three
dimensions to put a series of binary code in each region of how the brain
would look like.
-Ulysses

On 9/7/2017 5:22 AM, Piotr Machacz wrote:

I don't think LyreBird will ever be able to make local voices, SAPI or
otherwise. At least not in the near future. The way I understand it, their
system uses a massive GPU cluster to run a neural net that gets trained on
people's voices (this is why if you make enough recordings it learns to
mimic you so well). It literally makes connections and figures out patterns
like a human brain. If you were to run something like this on a single home
computer, even one with a beefy CPU or preferably a good high end GPU, I
imagine it would take days to train, and then minutes if not a few hours to
generate 1 clip of speech. That being said, neural nets are getting more and
more common and are started to be used on a small scale on computers or
phones, and some companies like Microsoft or Google are developing processor
chips designed specifically for neural nets. So maybe you'll see this become
an offline technology in a few years. For now, we know that lyrebird wants
to make an API available for this technology, so you can expect apps and
websites to make use of it (IE a chatting website might let you enter your
voice fingerprint and then you can talk to other people by typing text and
getting your actual voice out, or perhaps getting news or weather spoken to
you with your own voice)

On 7 Sep 2017, at 11:13, Jayson Smith <jaybird at bluegrasspals.com> wrote:

Hi,

I don't know of any other voice creation program which can make a SAPI
voice. If you still have the original recordings, and you still own the
copyright in and to them I.E. you didn't sign some sort of agreement giving
all your rights to Innoetics, you should in theory be able to use such a
service if one were to exist at some point in the future. You might try
contacting Lyrebird and explaining your situation, see what they say.

Sorry I can't be of more help,

Jayson

On 9/6/2017 3:03 PM, Blake Roberts wrote:
> Lyrebird sounds interesting. I notice that in FAQ Lyrebird says they

> can create a higher-quality voice if I have a lot of recordings but

> this is not available in the current beta. On a related topic, let me

> put all my cards on the table. Before Innoetics was acquired by

> Samsung, I was creating a SAPI of myself with help from a innoetics

> founder. In July 2017 around the time of Samsung's acquisition of the

> company, the Innoetics founder told me he would send me a SAPI of my

> voice using the thousands of sentences I recorded. Due to no SAPI

> received after almost 2 months from that promise, I'm thinking the

> Innoetics founder  whom I won't specify on-list might not be able to

> create Blake Sapi. I'm not trying to sound critical of the indivudal.

> I'm just accepting the possibility that aforementioned promise might

> not be fulfilled. Does anyone know of a voice creation program which

> can create a SAPI from already recorded sentence wave files? I spent

> many hours and months recording over a thousand  sentences for the

> aforementioned Innoetics Blake Sapi. I did the recording at no

> financial cost for personal/friends use. I don't want my time/effort

> to be wasted. Blake

> 

> -----Original Message----- From: Dectalk

> [mailto:dectalk-bounces at bluegrasspals.com] On Behalf Of Jayson Smith 

> Sent: Wednesday, September 06, 2017 12:20 PM To: DECtalk Discussions 

> Subject: [DECtalk] Lyrebird TTS, a demo is finally here!

> 

> Hi all,

> 

> 

> A few months ago, I heard about Lyrebird, a future project which

> would allow anyone to create a synthetic clone of their own voice.

> Last night I found out that it's finally here, in an early beta form.

> If you go to http://lyrebird.ai <http://lyrebird.ai/>  and create an
account, you can then

> record a minimum of thirty sentences they specify, the more you

> record the better, and then create your digital voice. Then you can

> have it speak any text you choose. I've played with it, and while the

> quality isn't the best, it does pretty accurately capture my voice,

> as well as most of the other people I know of who have created

> voices. Check it out!

> 

> 

> Jayson

> 

> 

> _______________________________________________ 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

> 

> 

_______________________________________________
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

_______________________________________________
Dectalk mailing list
Dectalk at bluegrasspals.com
http://bluegrasspals.com/mailman/listinfo/dectalk

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://bluegrasspals.com/pipermail/dectalk/attachments/20170907/70bd31bf/attachment-0001.html>