XHTML+Voice

~ the obit ~

On 21 December 2001 the W3C published XHTML+Voice Profile 1.0, a document written by Opera Software and IBM, with Håkon Wium Lie and T. V. Raman among the editors. The idea was that an ordinary web page could carry VoiceXML dialogs alongside its HTML, so a form could be filled in by typing or by speaking, and the page could read its own answers back to you. The usual shorthand for it was X+V.

The W3C's boilerplate on that document is unusually cold. Publication of a Note, it says, indicates no endorsement by W3C or the W3C Team, or any W3C Members. It goes on to state that no W3C resources were allocated to the issues addressed by the Note and that W3C had no editorial control over its preparation. Microsoft had filed a competing proposal the same year called SALT. Neither one ever became a Recommendation.

An implementation did arrive, which is more than most submissions get. Opera and IBM announced a joint multimodal browser in July 2002, Opera shipped voice support on Windows starting with version 8 in April 2005, and RFC 4374 registered the application/xv+xml media type in January 2006. For roughly seven years you could download a browser, install a voice package next to it and then talk to a web page.

~ the grammar you had to write ~

A page author wrote a grammar, which was a list of the things a user might plausibly say at that moment, and bound it to a form field. You pressed a key or said a trigger word, the speech engine matched what it heard against that list, and the matched value dropped into the input. The engine was IBM's embedded one, running locally on your machine. Nothing went to a server.

That last detail is also the thing that dated it. Because the grammar had to enumerate the possibilities in advance, X+V could only hear what the page had already imagined. The recognition quality was whatever shipped in the installer and it never got better on its own. Six years later the industry rebuilt the whole category around sending audio to a datacenter, and local enumerated grammars stopped looking like the future.

~ the rap sheet ~

~ the vendor stopped answering ~

Opera announced the cuts with its 12 beta on 25 April 2012. Voice went. Unite and Widgets went with it, for completely different reasons: those two were a strategy call about consolidating on one extensions platform.

The sentence about Voice was not a strategy call. It said that voice support would be removed as of Opera 12.00 because the voice-detection library was no longer supported by the third-party vendor. The third-party vendor was IBM, one of the two companies whose names are on the specification. Opera 12.00 went out on 14 June 2012 without it, and the only installable implementation of X+V in the world stopped existing that day.

The years when you were supposed to be talking to your browser are the years I spent on cafe machines in Turkey. Twenty of them in one room with Counter-Strike running on most, and headsets that belonged to the cafe rather than to anybody sitting there. Say a search query out loud in that room and you got looked at. Later, on a machine that was mine, Opera stayed the browser I read about rather than the one I ran, and the feature had about two years left by then.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »