- From: Deborah Dahl <Dahl@conversational-Technologies.com>
- Date: Thu, 5 Sep 2019 11:06:42 -0400
- To: <public-voiceinteraction@w3.org>
- Message-ID: <199301d563fb$872a86f0$957f94d0$@conversational-Technologies.com>
Here are the minutes from today's call. present: Dirk Schnelle-Walka, Debbie Dahl, Brian Susko chair: dirk scribe:ddahl meeting: voiceinteraction dirk: let's start with introductions, goals, process dirk: I was a researcher working on MMI architecture, focusing on voice, interested in architectural stuff brian: working on nlp space, conversations, interested in how IVA's talk to each other, how to send a payload to other IVA's ... getting different IVA's and IVR's to communicate debbie: VoiceXML and voice system, multimodal ...gave talk about standard NLP format at Voice Summit, there was a lot of interest debbie: note that minutes will be public dirk: what are our goals? ...IVA's could talk to each other, also how components could talk to each other brian: once IVA's get better and NLP gets more capable, routing between IVA's should be possible ...there should be a standard way of communicating payloads ...should be able to customize agents so you don't need to pull in a particular one dirk: accessing multiple IVA's in parallel brian: where is the inference engine for the intents? where does the responsibility lie for each function ...open up market to third party applications debbie: should there be a central registry of IVA's brian: yes, search IVA's by intention, there would be a gateway IVA ... that can be searched, like Maven dirk: we should aim for something like this debbie: details like where is the registry, how to describe the semantics of IVA's brian: there could also be some kind of certification, based on a training set ...that's probably down the road debbie: first we want to get the mechanisms in place and then think about certification dirk: there could be a central engine that selects IVA's brian: it could be like service integration, simpler than IVA integration dirk: cloud-based or also embedded? brian: cloud-based would be more extensible debbie: don't rule out embedded debbie: could either do a high level architecture or start with something specific brian: start out with architecture and then work on something detailed dirk: start with something general brian: high level diagram with initial focus dirk: can provide a high level diagram that we can discuss brian: we can see how that matches up with some of the ideas I've been thinking about debbie: possibly we can do a presentation at SpeechTEK debbie: people are seeing that the space is too fractured dirk: industry is starting to notice brian: would be good to bring this up in a public forum dirk: will set up a bi-weekly call for Wednesdays action: dirk to send proposal for high-level architecture for discussion next time
Received on Thursday, 5 September 2019 15:07:09 UTC