[DECtalk] new to the list

Wed Mar 1 00:58:10 EST 2006

>Hi Nick,
>Having been around DECtalk for a long time, my ears always perk up when
>someone says that most users prefer DECtalk for every day listening. Of
>course I believe you, but a lot of marketing and engineering folks say "so,
>where's the reference paper on that?" when I say I've heard that users
>prefer DECtalk. Has anyone ever seen one?

I've written one.  I can't share it for a number of reasons but my 
studies (of at least 77 participants) clearly show that formant 
speech like DECtalk is always preferred by screen reader users.

>
>OR -- maybe the more important question -- can you describe WHY you prefer
>DECtalk over NaturalVoices (you mean the ATT system, right? Not a natural
>human....?) - what listening conditions - or how listeners react
>differently?

Natural-sounding voices introduce a variable into the listening 
equation known as audio fatigue.  While the voices themselves sound 
very natural and pleasant in short bursts (one to two sentences), 
longer passages will begin to reveal certain audio artifacts 
(warbles, over and under emphasis, improper pitch contour, and 
different pronunciations of the same word depending on adjacent 
words).

All of these add up to audio fatigue and end up making the "natural" 
speech very frustrating for users.

What formant speech lacks in naturalness it more than makes up for in 
intelligibility.  A good formant synthesizer (like DECtalk) will 
pronounce things the same way at 500 words-per-minute as it does at 
200 words-per-minute.  Furthermore, words don't have different 
pronunciations depending on adjacent words (excluding homographs, of 
course).

Corporate marketing and the speech industry is not pushing formant 
speech because they don't believe this is what users want.  The 
appeal of demoing a human-sounding text-to-speech system is very sexy 
from a marketing perspective.  A formant speech system has no similar 
appeal, unfortunately.

One need only look as far as the "advancements" in DECtalk over the 
past five years.  In an attempt to make DECtalk sound more "natural," 
Force Computers and Fonix have both butchered a perfectly good 
formant text-to-speech synthesizer.  Development of advancements in 
formant TTS basically died with Dennis Klatt's passing; since then 
all companies have concentrated their efforts and R&D into improving 
the naturalness of speech.  As I illustrated earlier, this goal has 
yet to be met since audio fatigue is very real in AT&T Natural 
Voices--probably the most advanced natural-sounding TTS on the market 
today.

GUI Access

>
>
>-----Original Message-----
>From: dectalk-bounces at bluegrasspals.com
>[mailto:dectalk-bounces at bluegrasspals.com] On Behalf Of Nick G
>Sent: Monday, February 27, 2006 9:04 PM
>To: DECtalk Discussions
>Subject: Re: [DECtalk] new to the list
>
>Corine,
>If you are looking to get the sighted market on to it, and don't care about
>response time, then the natural voices is good.  However, most of us prefer
>things like DECtalk 4.40 for every day listening.
>Thanks,
>nick
>----- Original Message -----
>From: "Corine Bickley" <corine.bickley at gallaudet.edu>
>To: "'DECtalk Discussions'" <dectalk at bluegrasspals.com>
>Sent: Monday, February 27, 2006 10:44 AM
>Subject: Re: [DECtalk] new to the list
>
>
>>  Personally I would like
>>  to see a good, high quality open source synthesizer, either using
>>  already existing hardware such as the DEC-Talk or easily programmable
>>  software with good speech quality.
>>
>>  What programmable features would you like to see in an "easily
>>  programmable
>>  software" version?
>>  And, does "good speech quality" mean: 1. one of the older DECtalks (like
>>  4.40 or 4.3, NOT the new 5.0), or 2. some other kind of synthesizer, like
>>  ATT Natural Voices?
>>
>>  -----Original Message-----
>>  From: dectalk-bounces at bluegrasspals.com
>>  [mailto:dectalk-bounces at bluegrasspals.com] On Behalf Of Tony Baechler
>>  Sent: Monday, February 27, 2006 3:22 AM
>>  To: DECtalk Discussions
>>  Subject: Re: [DECtalk] new to the list
>  >
>>  Hi.  I'm sorry, but I'll have to disagree with you here.  First, the
>>  point of compiling a program in the first place is so that the source
>>  isn't easily apparent.  If it was possible to just randomly
>>  disassemble software, we wouldn't need the GPL, BSD, Linux or any
>>  other open source project.  I could just take the Windows kernel,
>>  crack it, invent my own version and release it as the free kernel or
>>  something.  It obviously doesn't work that way.  Also, remember a
>>  previous discussion about emulating the Echo on the Apple II?  The
>>  problem is the same in both cases.  The Echo is a chip on a card that
>>  goes into the computer.  The DEC-Talk is a separate
>>  computer.  According to the manual, it has its own
>>  microprocessor.  So, even if they had the firmware, it wouldn't do
>>  them much good.  It's a nice thought though.  Considering that it
>>  keeps getting sold, it's no surprise to me that they don't have the
>>  4.40 source, or won't release it if they do.  Personally I would like
>>  to see a good, high quality open source synthesizer, either using
>>  already existing hardware such as the DEC-Talk or easily programmable
>>  software with good speech quality.  I'm not interested in what's
>>  already out there for free, it all sounds like crap and won't sing
>>  very well if at all.
>>
>>  At 08:12 PM 2/26/2006, you wrote:
>>>I also don't understand how Fonix could have lost the code.  If the only
>>>copy is on a dead hard drive obviously they could send the drive in to a
>>>data recovery company if they still have it, but those places are
>>  expensive.
>>>I mean, with the move from Digital to Force and then from Force to Fonix,
>>>you'd think somebody somewhere would have something laying around.  I have
>>>the firmware version 4.2CD for the Express as well as the DECtalk PC
>>  drivers
>>>which as I understand it actually contain the DECtalk code which is loaded
>>>onto the board at startup.  And of course the 4.3 demo is available at the
>>>archive.  I don't know how to disassemble the software, but bet the right
>>>person who knew what they were doing could do it and create machine code
>>>from which new equivalent source code for the lost versions could be
>>>recreated.
>>
>>  _______________________________________________
>>  DECtalk mailing list
>>  DECtalk at bluegrasspals.com
>>  http://jaybird.no-ip.info/mailman/listinfo/dectalk
>>
>>
>>  _______________________________________________
>>  DECtalk mailing list
>>  DECtalk at bluegrasspals.com
>>  http://jaybird.no-ip.info/mailman/listinfo/dectalk
>>
>
>
>_______________________________________________
>DECtalk mailing list
>DECtalk at bluegrasspals.com
>http://jaybird.no-ip.info/mailman/listinfo/dectalk
>
>
>_______________________________________________
>DECtalk mailing list
>DECtalk at bluegrasspals.com
>http://jaybird.no-ip.info/mailman/listinfo/dectalk