Cepstral Community

Installed and licensed four languages: Callie, David, Marta & Millie
Installed but not yet licensed: Isabelle, Katrin, Vittoria

For reasons that are a mystery [to me], all TTS output comes with french license nag, even though I have licensed them.

The licenses said they installed OK. The clue that something is wrong is that when i check the version [cmd line: swift -v] it says the default voice is [french] Isabelle [i have no idea why], so I am thinking that since Isabelle is not licensed and she is the default, she sticks the nag in...then the text content has voice tags to switch it using a tag <voice name="David"> so i get the correct message in the correct voice...followed by the nag, again, by isabelle, in french.

so how do i fix this?

Interesting scenario.
What OS platform are you running on?
Can you provide a simple input to replicate this? (the exact input string with the embedded voice tag?)

My understanding:
- Isabelle is default and unlicensed
- TTS input does not call for Isabelle to speak, since a tag switches the voice to David immediately.
- The TTS output still has the french nag at the beginning.

Let me know if that interpretation is correct

UPDATE: to get into production, we did the $30 workaround - we purchased a license for Isabelle.

Your understanding of the setup is exactly correct. We are running a multi-lingual/multi voice large IVR where we actually are selecting voices dynamically based on a variety of dynamic conditions.

I am running on Centos 5.3 [Linux 2.6.18-92.1.22.el5].


A typical string:

<voice name="David-8kHz">I am sorry but the number you have entered does not match our records</voice>

This string was coming out exactly as expected but was [until above "workaround"] preceded by Isabelle's nag in french. To confirm, she was the unlicensed and the default.

From what I can see, the default seems to be based on the first language loaded, rather than the first language licensed. There was no obvious was to change the default that I could affect.

Thanks for looking into this.

We could mine this post for so many nice defects in specification and implementation, and we will. We will address the many issues appearing here in a subsequent release.

In the interim, (and honestly by looking through the source code) you can set the default voice on Unix using the environment variable SWIFT_DEFAULT_VOICE. (Please don't send Isabelle home...)

Not sure why this is not more prominent in the documentation.