SSML IPA support

[Tkael]

It looks like VoiceAttack may only support certain phoneme alphabets (unless I'm doing something wrong)? With SSML enabled, "Say Something with Text-to-Speech" should return "Pride of Achenar" for the string below, but instead it does not return any speech:

Code: [Select]

<phoneme alphabet="IPA" ph="praɪdɒvˈakɜːnɑ"> </phoneme>Thanks,


EDIT by Pfeil: Feature is already available

[Gary]

Try this example instead:

<phoneme alphabet="ipa" ph="praɪdɒvˈakɜːnɑ">Pride of Achenar</phoneme>


Note that, 'ipa' is lower-case (required) and that there needs to be something provided in the value (between the > and <).

[Gary]

Also, you'll need the full doc as well:

<?xml version="1.0" encoding="ISO-8859-1"?>
<speak version="1.0"
 xmlns="http://www.w3.org/2001/10/synthesis"
 xml:lang="en-US">

<phoneme alphabet="ipa" ph="praɪdɒvˈakɜːnɑ">Pride of Achenar</phoneme>

</speak>

[Tkael]

Thanks for your help Gary. I've found that VoiceAttack likes to crash whenever it encounters something it doesn't like in this area (such as a mismatch between the voice and xml:lang tag), but I've got a valid structure now. Unfortunately, it's still not handling IPA as I'd hoped. It is defaulting to the embedded text string rather than correctly parsing the IPA.

I'm using Ivona Amy, so I've set the xml:lang tag to `en-GB`.
The following should output "Pride of Achenar stands ready".
In Minispeech (TTS software that came with Ivona Amy) , I get the expected output while in VoiceAttack I instead get the output "Test stands ready".

Code: [Select]

<?xml version="1.0" encoding="ISO-8859-1"?>
<speak version="1.0"
 xmlns="http://www.w3.org/2001/10/synthesis"
 xml:lang="en-GB">

<phoneme alphabet="ipa" ph="praɪdɒvˈakɜːnɑ">Test</phoneme> stands ready.

</speak>


[Gary]

What SSML is crashing VA, exactly?

All VA is doing in regards to SSML is checking if the flag is set, and if the flag is set, send the text in its entirety to the TTS engine as SSML.  There's nothing going on other than that, so if there's a problem it's either going to be operator error or speech engine error (and unfortunately there's not much I'm going to be able to do about either unless I can spot a pattern).  In regards to the fault file, the latest entries seem to be emitted from interaction with EDDI.