SSML and 3rd party voices

[jcdick1]

If a voice is SAPI5 capable/compliant/whatever, I should be able to use SSML to tweak its pronunciation and such, correct?  If I use SSML in the TTS response to a command, the voice drops to the Microsoft voice, and sounds like Stephen Hawking.  The Ivona voice I'm using is supposed to be SAPI5, though.

Thoughts?  Corrections to my assumptions?

Thanks so much!

[Pfeil]

For anyone stumbling upon this topic: third-party TTS voices are likely to support SSML, however the SSML specification dictates that you use the "xml:lang" tag to specify which language the text will be read out in.

This means that if you're using a US English voice, you'd want to use a header like

Code: [Select]

<speak version="1.0" xml:lang="en-US">but if you're using a UK English voice, you'd want to use a header like

Code: [Select]

<speak version="1.0" xml:lang="en-UK">
If you, for example, set "xml:lang" to "en-US", if the voice you have selected in the "Voice" dropdown is not a US English voice, an available US English voice will be used instead of the selected voice.

This is not VoiceAttack-specific.

[Mike308]

Yes, I have had very good success implementing SSML into VA TTS responses. It is a bit of an art, but once you get the hang of it you can "performance direct" TTS output fairly reliably.