Re: Different behavior between RealSpeak voices
- From: Alex Midence <alex.midence@xxxxxxxxx>
- To: jfw@xxxxxxxxxxxxx
- Date: Thu, 12 Aug 2010 09:29:15 -0500
also, there's a new dectalk out. It's version 6. I have no idea
whatsoever if it has impacted the versions of the tts that are
available for use with screen readers but, who knows, they might've
fixed all the pronunciation bugs. Astoundingly good is an apt
description. It's pronunciation is clear, the responsiveness of the
speech engine is good, and, it can be sped up with no degradation of
the quality of the speech. Klatt deserves a Nobel Prize for his work,
IMHO. You can not get the same performance out of any concatonated
tts engine like the natural sounding voices. Here's why.
Warning: Technical stuff follows. Did my best to use plain English.
First, in case anyone is wondering, concatonated means strung
together. You do this in programming all the time. You can tell the
computer to have Q represent "Hello" and then fix it so that N stands
for name and leave it blank so someone can fill it in. You can then
tell the computer to ask for someone's name and tel it to display Q
plus n on the screen. The result is "Hello Fred" or "Hello Mary" or
whatever you put in. You had it concatonate hello and fred to make
that sentence. Same principal is applied with natural voices both the
ones free with Jaws and the others you buy like Cepstral and NeoSpeech
and all of them. It basically means that they have a bunch of
clusters of sounds and sylables called phonemes that they string
together like beads on a necklace to make the synthesizer make a word
or sentence. They sometimes even go so far as to use a real human
voice, digitize the individual phonemes produced when that person
speaks and hope that when those phonemes (which are stored in some
sort of data base of sound files are strung together during the
reading of text by the machine, that they will be as close to the real
person's speech as possible. The natural voices have large chunks of
these phonemes with included pauses that make up a unit that is then
used for that sort of stringing together. Traditional speech
synthesizers use a more digital approach. They don't concern
themselves with natural human patterns of speech or breathing or any
of that. It's all numbers and units of time. Make that vowel x
milliseconds long, make that s or t or f have such and such a
frequency, pause for x number of milliseconds after commas, or periods
or semicolons; that sort of thing. When you speed the speech up, it
just scales those intervals to the speed of the speech you have it set
to. The concatonated tts's are still too busy trying to mimic natural
rhythms of speech and preserve the fidelity of the human quality of
the voice so, you get this little shiver or quaver at the beginning of
some sylables because, frankly, people just don't talk that fast. You
get the reduced responsiveness because complex sound files like that
take up lots of memory more so than the digital artificial rendering
of sound you get from the more robotic voices. Those have been around
for so long that they were originally used in pc's like 8086's (I
learned Dos on such a machine using Vert, anyone remember Vert? Oh,
the pain! Oh, the suffering!). Those pc's had tiny ram memories like
640k (not even a meg, folks) and they worked just fine with those
synthesizers. You could even run DecTalk on those machines (Vocal
Eyes, anyone?), and still have good responsiveness.
Sorry about the length. I'm a linguist/programmer/technoNerd and this
really got me going. Any corrections to what I have stated above are
welcome. a lot of this is the product of curiosity-driven reading and
a bit of educated hypothesis.
Alex M
On 8/12/10, Adrian Spratt <Adrian@xxxxxxxxxxxxxxxx> wrote:
> Ha. You're right about that example, which is easily fixed through the JAWS
> dictionary. But for the most part, Dectalk Express pronunciation is
> astoundingly good.
>
> -----Original Message-----
> From: jfw-bounce@xxxxxxxxxxxxx [mailto:jfw-bounce@xxxxxxxxxxxxx] On Behalf
> Of Kerri
> Sent: Wednesday, August 11, 2010 11:31 AM
> To: jfw@xxxxxxxxxxxxx
> Subject: Re: Different behavior between RealSpeak voices
>
> Dectalk is superior to most synthesizers out there but it does have a lot of
> pronunciation difficulties. It says "strung" instead of "stung"
> ----- Original Message -----
> From: "Alex Midence" <alex.midence@xxxxxxxxx>
> To: <jfw@xxxxxxxxxxxxx>
> Sent: Wednesday, August 11, 2010 7:13 AM
> Subject: Re: Different behavior between RealSpeak voices
>
>
>> I'm partial to Huge Harry, myself. I have him at 425 words per minute
>> on my Reading Edge with inflection set to highest and it sounds just
>> fine to me. Wife thinks it sounds like an auctioneer on speed.
>> Sounds just fine to me though. I never understood why people are down
>> on artificial voices. If you're using a computer, why do you expect
>> it to sound human? Persoanlly, it creeps me out a bit.
>>
>> Alex M
>>
>> On 8/11/10, Lisle, Ted (CHFS DMS) <Ted.Lisle@xxxxxx> wrote:
>>> I find I have to listen much harder, and somewhat slower, to an
>>> external voice. I can handle Glenn at a tremendous pace, and never miss a
> word.
>>> As fast as current hardware busses are, it simply may not be possible
>>> to get the kind of tight integration and smooth phrasing from an
>>> external voice that we can from one incorporated into the screenreader.
>>>
>>> Ted
>>>
>>> -----Original Message-----
>>> From: jfw-bounce@xxxxxxxxxxxxx [mailto:jfw-bounce@xxxxxxxxxxxxx] On
>>> Behalf Of Yadiel
>>> Sent: Tuesday, August 10, 2010 9:37 PM
>>> To: jfw@xxxxxxxxxxxxx
>>> Subject: Re: Different behavior between RealSpeak voices
>>>
>>> Yes! Although I prefer jill. "She" sounds... well... in lack of a
>>> better
>>>
>>> phrase, more sweet and plecent to the ear. Although, not as responsive.
>>> In
>>> fact, I yet haven't found a voice that is as responsive as I would like.
>>>
>>> Real Speak Solo direct or sappy 5, I still can't find a good sounding
>>> responsive voice. That is why I use eloquence for most of my task
>>> including web surfing and use the real speak and sappy 5s for reading
>>> very long documents.
>>>
>>>
>>> Yadiel
>>>
>>> --------------------------------------------------
>>> From: "Dale E. Heltzer" <deheltzer@xxxxxxx>
>>> Sent: Tuesday, August 10, 2010 8:47 PM
>>> To: <jfw@xxxxxxxxxxxxx>
>>> Subject: Different behavior between RealSpeak voices
>>>
>>>> Some days ago I posted about how RealSpeak Daniel doesn't always
>>>> pronounce words the way they're written.
>>>> I've been using Samantha, and have noticed no such shortcomings -
>>>> and "she" is much more responsive than Daniel.
>>>>
>>>> Anyone else notice this?
>>>> ---
>>>> Dale E. Heltzer
>>>> deheltzer@xxxxxxx
>>>>
>>>>
>>>> --
>>>> JFW related links:
>>>> JFW homepage: http://www.freedomscientific.com/ Scripting mailing
>>>> list:
>>>> http://lists.the-jdh.com/listinfo.cgi/scriptography-the-jdh.com
>>>> JFW List instructions:
>>>> To post a message to the list, send it to jfw@xxxxxxxxxxxxx To
>>>> unsubscribe from this mailing list, send a message to
>>>> jfw-request@xxxxxxxxxxxxx with the word unsubscribe in the subject
>>> line.
>>>> Archives located at: //www.freelists.org/archives/jfw
>>>> Alternative archives located at:
>>>> http://n2.nabble.com/JAWS-for-Windows-f2145279.html
>>>>
>>>> If you have any concerns about the list, post received from the
>>>> list,
>>> or
>>>> the way the list is being run, do not post them to the list. Rather
>>>> contact the list owner at jfw-admins@xxxxxxxxxxxxxx
>>>>
>>> --
>>> JFW related links:
>>> JFW homepage: http://www.freedomscientific.com/ Scripting mailing
>>> list:
>>> http://lists.the-jdh.com/listinfo.cgi/scriptography-the-jdh.com
>>> JFW List instructions:
>>> To post a message to the list, send it to jfw@xxxxxxxxxxxxx To
>>> unsubscribe from this mailing list, send a message to
>>> jfw-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.
>>> Archives located at: //www.freelists.org/archives/jfw
>>> Alternative archives located at:
>>> http://n2.nabble.com/JAWS-for-Windows-f2145279.html
>>>
>>> If you have any concerns about the list, post received from the list,
>>> or the way the list is being run, do not post them to the list.
>>> Rather contact the list owner at jfw-admins@xxxxxxxxxxxxxx
>>> --
>>> JFW related links:
>>> JFW homepage: http://www.freedomscientific.com/ Scripting mailing
>>> list:
>>> http://lists.the-jdh.com/listinfo.cgi/scriptography-the-jdh.com
>>> JFW List instructions:
>>> To post a message to the list, send it to jfw@xxxxxxxxxxxxx To
>>> unsubscribe from this mailing list, send a message to
>>> jfw-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.
>>> Archives located at: //www.freelists.org/archives/jfw
>>> Alternative archives located at:
>>> http://n2.nabble.com/JAWS-for-Windows-f2145279.html
>>>
>>> If you have any concerns about the list, post received from the list,
>>> or the way the list is being run, do not post them to the list.
>>> Rather contact the list owner at jfw-admins@xxxxxxxxxxxxxx
>>>
>> --
>> JFW related links:
>> JFW homepage: http://www.freedomscientific.com/ Scripting mailing
>> list:
>> http://lists.the-jdh.com/listinfo.cgi/scriptography-the-jdh.com
>> JFW List instructions:
>> To post a message to the list, send it to jfw@xxxxxxxxxxxxx To
>> unsubscribe from this mailing list, send a message to
>> jfw-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.
>> Archives located at: //www.freelists.org/archives/jfw
>> Alternative archives located at:
>> http://n2.nabble.com/JAWS-for-Windows-f2145279.html
>>
>> If you have any concerns about the list, post received from the list,
>> or the way the list is being run, do not post them to the list. Rather
>> contact the list owner at jfw-admins@xxxxxxxxxxxxxx
>
> --
> JFW related links:
> JFW homepage: http://www.freedomscientific.com/ Scripting mailing list:
> http://lists.the-jdh.com/listinfo.cgi/scriptography-the-jdh.com
> JFW List instructions:
> To post a message to the list, send it to jfw@xxxxxxxxxxxxx To unsubscribe
> from this mailing list, send a message to jfw-request@xxxxxxxxxxxxx with the
> word unsubscribe in the subject line.
> Archives located at: //www.freelists.org/archives/jfw
> Alternative archives located at:
> http://n2.nabble.com/JAWS-for-Windows-f2145279.html
>
> If you have any concerns about the list, post received from the list, or the
> way the list is being run, do not post them to the list. Rather contact the
> list owner at jfw-admins@xxxxxxxxxxxxxx
>
> --
> JFW related links:
> JFW homepage: http://www.freedomscientific.com/
> Scripting mailing list:
> http://lists.the-jdh.com/listinfo.cgi/scriptography-the-jdh.com
> JFW List instructions:
> To post a message to the list, send it to jfw@xxxxxxxxxxxxx
> To unsubscribe from this mailing list, send a message to
> jfw-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.
> Archives located at: //www.freelists.org/archives/jfw
> Alternative archives located at:
> http://n2.nabble.com/JAWS-for-Windows-f2145279.html
>
> If you have any concerns about the list, post received from the list, or the
> way the list is being run, do not post them to the list. Rather contact the
> list owner at jfw-admins@xxxxxxxxxxxxxx
>
--
JFW related links:
JFW homepage: http://www.freedomscientific.com/
Scripting mailing list:
http://lists.the-jdh.com/listinfo.cgi/scriptography-the-jdh.com
JFW List instructions:
To post a message to the list, send it to jfw@xxxxxxxxxxxxx
To unsubscribe from this mailing list, send a message to
jfw-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.
Archives located at: //www.freelists.org/archives/jfw
Alternative archives located at:
http://n2.nabble.com/JAWS-for-Windows-f2145279.html
If you have any concerns about the list, post received from the list, or the
way the list is being run, do not post them to the list. Rather contact the
list owner at jfw-admins@xxxxxxxxxxxxxx
Other related posts:
