RE: [css-speech] Heads-up: CSS WG plans last call for css3-speech (part 2)

Thank you for your kind consideration of my comment.

I understand that the CSS Working Group has unfortunately chosen to defer this issue and that, as such, is deferring the ability to specify a voice family with a particular 'accent' (e.g. a French-accent voice to read an English film title). I will look forward to seeing this in the next iteration of CSS-Speech.

With regards,
Paul

-----Original Message-----
From: Daniel Weck [mailto:daniel.weck@gmail.com] 
Sent: Friday, October 14, 2011 12:00 PM
To: BAGSHAW Paul RD-TECH-REN; www-style list; w3c-voice-wg@w3.org
Subject: Re: [css-speech] Heads-up: CSS WG plans last call for css3-speech (part 2)

Dear Paul (et al),
the CSS Working Group approved [1] the decision to defer this issue to a later revision of the CSS Speech Module, due to the technical considerations mentioned during our email discussion (partially quoted below).

However, it is clear that the statement about the relationship between the CSS Speech Module Level 3 and SSML1.1 needs to be clarified, we will therefore implement the required (minor) editorial changes.

Please let us know whether you are in support of the WG consensus.
Kind regards, Daniel

[1]
http://lists.w3.org/Archives/Public/www-style/2011Oct/0482.html


On 8 Sep 2011, at 17:07, Daniel Weck wrote:
> SSML1.1 normatively references BCP47, which defines not only the syntax for language identifiers (used in xml:lang [1] and in the 'languages' attribute [2] of SSML markup), but also the language matching algorithm. Furthermore, the voice selection algorithm in SSML 1.1 relies on the "voice descriptions" [3] and "onlangfailure" [4] conformance requirements ('must') for speech processors.
> 
> CSS Level 3 supports a looser language model [5] (which doesn't normatively reference BCP47), and the Speech Module relies on a number of implementation-dependent properties of the underlying speech processor [6] (i.e. areas where there is no such strict conformance requirements as the ones in SSML). Although we aim at matching most SSML features, there is no 1-to-1 mapping due to intrinsic technical differences between SSML versus CSS document models.


On 29 Sep 2011, at 10:02, Daniel Weck wrote:
> in my reply, I only mentioned 'xml:lang' in between parenthesis, to clarify the context within which BCP47 is referenced (in SSML1.1), and to draw a parallel with CSS3-Speech. Rest assured that I do not find the semantics of 'languages' confounding. Note that I brought-up SSML's 'onlangfailure' feature because CSS-Speech would need to normatively document conformance requirements for synthesis processor ("voice descriptions", BCP47, etc.).
> 
> Given that the 'languages' attribute is optional in SSML1.1, why do you think that "the ability to generate a conformant SSML 1.1 document through application of a CSS Speech Model" would be "severely hindered"? I appreciate that round-trip engineering would be an issue (as with other features in CSS3-Speech), but the CSS -> SSML translation would seem to work fine.


> 
> On 18 Aug 2011, at 12:00, <paul.bagshaw@orange-ftgroup.com> <paul.bagshaw@orange-ftgroup.com> wrote:
>> 
>> 2. <languages> value need for voice selection in 'voice-family' property.
>> 
>> SSML 1.1 has made a significant move from SSML 1.0 in its voice selection algorithm. It has brought a much needed clarification between the language of written content and the language(s) spoken by a particular voice (to allow, for example, a French voice to read an English film title). The xml:lang attribute is used by SSML 1.1 only to designate the language of written content. The SSML 1.1 <voice> element takes an optional 'languages' attribute "indicating the list of languages the voice is desired to speak".
>> 
>> It is desirable to give access to this attribute via the 'voice-family' property.
>> 
>> The voice selection algorithm proposed by the CSS-Speech module should consequently to put in line with that specified by SSML 1.1.

Received on Friday, 14 October 2011 11:07:23 UTC