[asterisk-users] cepstral vs festival

[John Todd]

On Dec 2, 2008, at 9:41 AM, Erik (Caneris) wrote:

> Festival sucks. Cepstral sucks less. The End.
>
> In my experience, it depends on the specific app, who's paying, and  
> who's going to be the victim, err...user listening to it. This is  
> the difference between domain/context specific phrases/words to  
> pronounce vs. general stuff, a client on a tight budget or not, the  
> users being internal vs. customers/public, and so on.
>
> Cepstral is a $30 TTS engine. It's not too bad, but you'll find  
> mostly things like Realspeak deployed in large scale "professional"  
> deployments, such as those used by the "big boys", telcos/banks/ 
> airlines. We deployed Cepstral recently for a client, for a phone-in  
> service used by the general public, and I can tell you that there  
> was quite a bit of work in "teaching" it with SSML how to pronounce  
> stuff.
>
> Again, it really depends on your specific situation. You should  
> definitely try out those two at least and also ensure that the  
> client/stakeholders are aware of limitations. There's a certain  
> expectation of "it will speak perfectly" these days, followed by  
> disappointment and blame when reality hits them.
>
> Regards,
> --
> Erik
> Caneris
> Tel: 647-723-6365
> Fax: 647-723-5365
> Toll-free: 1-866-827-0021
> www.caneris.com


Erik -
   Have you found RealSpeak to be worth the cost?  Can Cepstral, with  
the hourly $ spent on tuning, be made to be a reasonable substitute?   
It's been a while since I did a head-to-head comparison between  
Cepstral and (anything else) so I did a quick demo of the RealSpeak  
Host-based telecom app:

   http://www.nuance.com/realspeak/demo/  (contact data required)

and the Cepstral demo:

   http://www.cepstral.com/demos/

I used the "Jill (default - 8khz)" for RealSpeak and "Allison  
(default)" for the tests, and played back the same phrase:

   "Congratulations. You have successfully installed and executed the  
Asterisk open source PBX."

My results: The RealSpeak sample was more clear than the Cepstral.   
But by how much?  I should probably test with more than just that one  
phrase, but I can't say I'd prefer RealSpeak significantly over  
Cepstral in this extremely limited case.  Does RealSpeak get better  
long-term test results and comprehension/retention?  I know that  
Cepstral is $50/port - the RealSpeak pricing is un-findable, which  
tells me that it's significantly higher than Cepstral.  (Personal  
peeve: at least put your list pricing on the website! <grumble>)

That being said, I'd really be interested in hearing if anyone has  
done a RealSpeak-to-Asterisk conduit.  I wasn't able to quickly  
uncover how they interact with third-party systems - is it VoIP?  A C  
library?  Some sort of HTTP socket?  The more methods we can get  
working with Asterisk, the better, because not every implementation of  
a voice system has the same requirements...

JT

---
John Todd
jtodd at digium.com        +1-256-428-6083
Asterisk Open Source Community Director