- From: JOHNSTON, MICHAEL J (MICHAEL J) <johnston@research.att.com>
- Date: Wed, 29 Jun 2011 20:00:25 -0400
- To: Robert Brown <Robert.Brown@microsoft.com>
- CC: HTML Speech XG <public-xg-htmlspeech@w3.org>
The more I look at what we are proposing for extensibiity beyond speech transcription, the more it looks we really are shoe-horning too much into the recognize method. You end up with too much introspection being required at the server side. On receiving a set-grammar request the service has to inspect the model in order to work out whether it is a speech recognition model or something else. As an alternative we could start from a generic 'control' method with a control type e.g. 'RECOGNIZE' 'SET-GRAMMAR' .... where those are drawn initially from MRCP2, but also future proof the protocol by allowing application and vendor specific control types. This simplifies matters for developers since 'RECOGNIZE' will really mean do speech recognition and 'SET-GRAMMAR' will really mean set the speech recognition grammar ... On Jun 24, 2011, at 9:58 PM, Robert Brown wrote: It’s a pretty rough draft, but I _think_ this captures everything we have so far. (Keep me honest - if I missed something, let me know). I’ve inserted a number of “TODO” items, which are either open issues we’ve discussed but haven’t closed on, or questions or work items that occurred to me as I compiled this draft. Have a good weekend. Cheers, /Rob <speech-protocol-basic-approach-02.html>
Received on Thursday, 30 June 2011 00:00:52 UTC