W3C home > Mailing lists > Public > www-voice@w3.org > October to December 2003

RE: Question about using TTS via the prompt element in VXML

From: Brian Wyld <brian.wyld@eloquant.com>
Date: Thu, 4 Dec 2003 10:02:14 +0100
To: "Tracy Boehrer" <tboehrer@calltower.com>, "Lutz Birkhahn" <lbirkhahn@adomo.com>, "Roopa Trivedi" <rotrived@cisco.com>, <www-voice@w3.org>
Message-ID: <OCENLGFFCDHPOEENHEPFMELAELAA.brian.wyld@eloquant.com>


I think Lutz's is correct point of view - if there is a voice named, then use it, else use the 'characteristics' of the voice to select the best match.

One implementations point of view: (:-)
 - find TTS engine by language first (per service we allow only one engine type per locale, potentially on multiple servers) -> this makes sure that at least the voice selected has the right accent....
 - if voice name given, and exists, use it.
 - else, from list of all possible voices for the language, select by the ordered list { gender, age, varient}

Hence the selection algo uses the list in the order given in the spec.....
However it is not possible at the SSML level to select a specific vendor's engine.....


[Brian Wyld] [brian.wyld@eloquant.com]
[Directeur General R&D]
[Eloquant SA] [+33 476 77 46 92] [www.eloquant.com]
[advanced solutions for telecoms and IT services] 

> -----Message d'origine-----
> De : www-voice-request@w3.org [mailto:www-voice-request@w3.org]De la
> part de Tracy Boehrer
> Envoyé : Wednesday, December 03, 2003 23:51
> À : Lutz Birkhahn; Roopa Trivedi; www-voice@w3.org
> Objet : RE: Question about using TTS via the prompt element in VXML
> The SSML spec pretty much says that (section 2.2.1).  It even 
> states what attributes to use.  But I don't think it's clear 
> enough about the algorithm. In fact, it says "the voice selection 
> algorithm may be processor specific."
> I would think it needs to be ordered.  For example: language, 
> gender, age, variant, name.  Or something like that.  Or, is the 
> order the spec lists it the right way: language, name, variant, 
> gender and age?
> I don't know...
> 	-----Original Message----- 
> 	From: Lutz Birkhahn [mailto:lbirkhahn@adomo.com] 
> 	Sent: Wed 12/3/2003 2:14 PM 
> 	To: Roopa Trivedi; Tracy Boehrer; www-voice@w3.org 
> 	Cc: 
> 	Subject: Re: Question about using TTS via the prompt element in VXML
> 	Roopa Trivedi wrote:
> 	> What about the scenario where 2 vendors support the same 
> gender, age,
> 	> language etc and the script writer specifically wants to 
> use vendor #1?
> 	> Is the "name" of the voice recommended to be used as the 
> distinguishing
> 	> factor? Can we assume that this "name" will be unique across all
> 	> vendors?
> 	I think voices in VXML are what fonts are in HTML (or TeX, or Post-
> 	Script)... Maybe that's a good analogy when discussing this 
> question.
> 	From that point of view, "name" of a voice is a possible 
> way to select
> 	a voice, but that name should include the vendor ("foundry") to make
> 	it unique ("Scansoft-Jim", "Viavoice-Rebecca"). One 
> difference is that
> 	font names are usually protected, so when I ask for a Times 
> font, I can
> 	expect to get something which looks at least somewhat similar to any
> 	other Times font. I don't think the same holds for a voice 
> called "Jim".
> 	ScanSoft's Jim might sound very different from IBM's Jim (although
> 	both are probably male american voices).
> 	In general it would be nice to select a voice depending on 
> any mixture
> 	of attributes, e.g. by specifying *some* of the attributes gender,
> 	age(?), language, country, name, vendor, etc., much the same as e.g.
> 	in X Window System I can specify a font similar to "*-Times-bold-*"
> 	when I don't care where the Times font comes from, or I can specify
> 	"Adobe-Times-bold-*" when I insist on having the Adobe 
> voice err font.
> 	Just my $.02,
> 	/lutz
> 	--
> 	Lutz Birkhahn                                       System 
> Software Engineer
> 	Adomo Inc. --  10001 N. De Anza Boulevard Suite 220  --  
> Cupertino, CA 95014
Received on Thursday, 4 December 2003 04:03:15 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 21:07:37 UTC