[ILUG] Re: Text to Speech

[Conor McCarthy]

hey all,
 
> There are a few good utils available.  Probably the one everyone is going to
> tell you about is Festival, but will require a lot of RPM's, a lot of
[...]
 
Far too much singing and dancing for a speech program. And far too much lisp.

I use mbrola, which is technically a diphone synthesis program, not a TTS.
A diphone is a "speech component", you can't just type in English and have
it read it out to you. You do type in the diphones, duration and pitch data.
Or better still, get freephone which uses a 10k word dictionary (electronic 
version of the Oxford English pronounciation dictionary) which outputs the 
necessary. It uses GNU dbm to compile the data, and it's easy to add your
own words (it's only 99% robust, it fails to produce valid data causing
mbrola to not produce output for certain inputs, there are diphone 
combinations which are "not valid" in speech).

Also, each language has slightly different diphones, mbrola has several
diphone databases, US/UK-English, Spanish, German, French and more.
You can even make it sing and yodel :)

They seem to have added plenty of extra stuff since I last looked at the 
site, so it's probably a lot easier to use now.
<URL:http://tcts.fpms.ac.be/synthesis/mbrola.html>

> PS:  Is there a good reason why we cant post to this board using Netscape
> while logged in as God, *cough* root?  I dont fancy reverting to a user
> account, even if it is a good idea.

shudder.

C.
--
Conor McCarthy               Grabel's Law: 2 is not equal to 3, not even
Computing Services                           for very large values of 2.
University College Dublin            <URL:http://lurch.ucd.ie/~conormc/>
Ireland                    Phone: +353-1-706-2456   Fax: +353-1-283-7077