Re: [css3-speech] Test suite

It's an interesting question. If I had to regression test an implementation to make sure no bugs were introduced between versions I'd be looking to output text files for comparison. The synthesizer I know is Apple's so probably I'd start by getting the synthesizer to dump the phonemes it will generate. Then I there's another format it can create that is far more detailled (IIRC it is called TUNE) that I believe would show up things like pitch and volume changes in the text output. 

But those are proprietary and thus useless for comparing implentations (well other than 2 UAs using the same underlying synthesizer) and entirely inappropriate for a CSS test suite. 

Could SSML help here? Given the close correspondence of css3 speech with SSML could one form test cases which say "this HTML + this CSS = this SSML"

Does that actually help? It rather assumes that

i) implementations would be willing to generate SSML as an alternative to calling a synthesizer to generate speech

ii) given an implementation that generates a correct SSML document for some input then when the same implementation switches to generating speech the output is equivalent. 

If there were a reference speech synthesizer that took SSML as input and was known to produce 'correct output' this would be an easier sell. 

What I'm trying to do obviously is decompose the problem into smaller ones that are more amenable to automation. Unfortunately I don't think the necessary pieces all exist. 

On Jun 29, 2011, at 11:38 PM, Simon Fraser <smfr@me.com> wrote:

> I'm curious about what the test suite for CSS3 speech would be like. Would it involve someone having to listen through several hundred test cases? Is there some other, more mechanized way to do the testing?
> 
> Simon
> 
>

Received on Thursday, 30 June 2011 13:14:57 UTC