[access-uk] Re: Acapela Voices.
- From: Léonie Watson <tink@xxxxxxxxxx>
- To: <access-uk@xxxxxxxxxxxxx>
- Date: Sat, 17 Jan 2009 16:52:33 -0000
Glad you found it useful. I agree about the use of concatonative
synthesis.
It's a funny thing that people are usually willing to suspend their
belief completely, or not at all. If something sounds utterly unrealistic,
people tend to accept it for what it is. If something is close to being
realistic, but not quite, the affect tends to be much more jarring.
Léonie.
--
http://www.tink.co.uk/
_____
From: access-uk@xxxxxxxxxxxxx [mailto:access-uk@xxxxxxxxxxxxx] On Behalf Of
Mike Cassidy
Sent: 15 January 2009 15:39
To: access-uk@xxxxxxxxxxxxx
Subject: [access-uk] Re: Acapela Voices.
Hi Tink,
This is a late reply, but what a very helpful comment. Funnily enough, I
prefer the Eloquence-type synthesis; I find the concatenative speech almost
over-real, just too exaggerated.
Thanks,
Mike
-----Original Message-----
From: access-uk@xxxxxxxxxxxxx [mailto:access-uk@xxxxxxxxxxxxx] On Behalf Of
Léonie Watson
Sent: 20 October 2008 17:27
To: access-uk@xxxxxxxxxxxxx
Subject: [access-uk] Re: Acapela Voices.
It may not be the case here, but it's worth knowing that the Acapela
voices use concatonative speech synthesis, as opposed to engines such as
Eloquence, witch use formative speech synthesis. The trade off is between
realism and responsiveness.
Concatonative synthesis is based on recordings of real voices. Acapela
use a form of concatonative synthesis known as unit selection synthesis.
Essentially, this breaks down the real voice recordings into phones,
syllables, words and phrases and stores them in a database. When text is
converted into speech, the text is analysed and the resulting speech is
built from the building blocks held in the database.
Formative synthesis relies entirely on artificially generated sound. The
upshot is that although concatonative synthesis sounds more real, it is
often less responsive and prone to technical glitches. Formative synthesis
sounds much less real, but tends to be more responsive.
Léonie.
--
http://www.tink.co.uk/
_____
From: access-uk@xxxxxxxxxxxxx [mailto:access-uk@xxxxxxxxxxxxx] On Behalf Of
Roy Bannister
Sent: 20 October 2008 09:41
To: access-uk@xxxxxxxxxxxxx
Subject: [access-uk] Acapela Voices.
Thanks to Steve sending a demo disk of Acapela voices I have had a chance to
try odd ones. I normally use Reed from Window-Eyes Eloquence, or Paul on
Satogo Dectalk. I found Peter and Ryan were acceptable for reading text
passages at a reasonable speed, found them slightly jerky with final
syllables sometimes clipped. However when typing using either of them I find
them rather sluggish, also the upper case letters are not announced in a
higher pitch, I definitely don't care for this. I have Skread installed on
my machine with Skype, using my normal voices, the Skread voice will
announce details whilst the other voice may be reading something else, this
doesn't happen with the Acapela voices, they are not trying to use the same
voice. My machine is a dual core 2.8Gb processor with 1Gb of ram, this makes
me suspect that the Acapela voices are memory hungry. I will be interested
in other users verdicts.
Cheers
Roy
roy.bannister@xxxxxxxxxxxx
Skype: roybannister4787
Other related posts:
