[asterisk-biz] Pre-recorded first and last names audio database

[Lee Jenkins]

John Vogel wrote:
> Thanks for the suggestion, Dean
> 
> I've tried ATT Natural Voices
> (http://www.research.att.com/~ttsweb/tts/demo.php) and Microsoft Speech
> Server just to see how the current state-of-the-art TTS performs. They do
> fine with easy names like "John" and "Smith" but even something as simple as
> "Martinez" gets mangled (accent on the wrong syllable).
> 
> In general TTS seems to work better on first names (which tend to be
> shorter) then last names.
> 
> Regards,
> John
> 
> 

John,

I've been using Cepstal quite a bit personally and the pronunciations 
seems to be pretty decent.  (Using mostly the "William" voice.)

Using either odd spellings or better yet, using SSML 
(http://www.w3.org/TR/speech-synthesis/) you can fine tune the 
pronunciations quite a bit.

I wrote an AGI for cepstral a while back that you can use (or you can 
call its functions directly from the dialplan).

http://www.voip-info.org/wiki/view/DTSwift+Cepstral+AGI+Wrapper

It's written with Freepascal and the source is available from the page 
above.  I have an i386 binary available as well if you don't want to 
compile it (email me privately for it).

I also added an class for Cepstral to the AsterPas FastAGI Script Server 
(http://www.datatrakpos.com/pos/datatalk/asterpas.aspx) if you're into 
pascal programming.


--
Warm Regards,

Lee