15:02:14 RRSAgent has joined #voiceinteraction 15:02:18 logging to https://www.w3.org/2024/01/10-voiceinteraction-irc 15:02:31 meeting: Voice Interaction 15:02:40 chair: debbie 15:02:56 scribe: ddahl 15:04:47 agenda: https://lists.w3.org/Archives/Public/public-voiceinteraction/2024Jan/ 15:33:08 present: dirk, debbie, gerard 15:33:26 topic: dirk's implementation 15:34:19 dirk: no client yet 15:34:31 ...ipa registry 15:34:53 ...ipa provider is chatGPT 15:35:29 ...using client input, sends text to dialog layer 15:35:53 ...the concepts are working 15:36:20 ...creates provider selection service 15:36:37 implementation is C++ 15:37:28 dirk:only uses "multimodal input" parameter 15:37:43 ...this is a container of various multimodal inputs 15:38:50 ...returns client response 15:39:33 ...session id, request id, multimodal output 15:42:37 ...the ipa service must know about the provider selection service 15:44:12 ...provider selection service knows about the provider registry 15:44:32 ...provide registry is hard-coded, knows about providers 15:46:33 ...chatGPT is one provider 15:50:17 ...uses provider selection strategy, which hasn't been implemented yet, just uses the first one 15:53:27 ...eventually will need to use keyboard and display in client 15:56:40 ...need to add more logic about how to select IPA 15:59:31 debbie: seems like picking the IPA is the job of the provider selection service 15:59:39 ...not the registry 16:00:52 ...location of chatGPT is part of the adapter, not the registry 16:02:17 debbie: could use a similar pattern for Amazon Lex, for example 16:02:53 dirk: provider selection strategy should know about supported modalities 16:03:38 ...could try to match input with output 16:03:58 ... adapter should mention what it supports 16:04:12 ...in terms of modalities 16:04:25 ...we never discussed error situations 16:05:08 ...for example, ChatGPT gave an error for incorrect credentials 16:06:12 dirk: should return exceptions if the input is bad 16:06:43 ...rather than error codes 16:07:37 ... we need to decide this strategy soon 16:08:40 ...this is in the repository, where to put the documentation 16:09:41 debbie: should be in the "source code folder 16:12:56 dirk: what we've defined so far seems to be working 16:13:35 debbie: next priority is provider selection service and registry need to be fleshed out 16:14:02 ...maybe the client isn't that critical 16:14:15 dirk: would need that if we don't want to recompile every time 16:14:55 dirk: need to check license of source code 16:15:22 debbie: check to see if w3c has a preference 16:15:59 ...could use Apache 2.0 if the W3C doesn't have a preference 16:16:15 dirk: then elaborate on provider selection strategy 16:16:27 ...decide on return values vs. exceptions 16:16:56 debbie: providers should advertise what modalities they support 16:17:13 dirk: this would be part of the provider selection strategy 16:18:16 ...highest priority is to document in the README 16:18:31 ...invite others to support 16:19:09 dirk: still planning to look at dialog events 16:19:56 debbie: should also look at conversation envelope spec, which will be published in the next few days 16:20:36 dirk: didn't run into any major problems with implementation 16:21:41 dirk: should create issues for next steps 16:22:57 ...any questions from gerard? security? 16:23:37 gerard: implementation of small LLM's on smartphone 16:24:32 ...European project called "Evita", coaching older people 16:24:56 ...want to avoid cloud access 16:25:34 dirk: would it make sense to add this as a wrapper around your project? 16:25:38 gerard: not sure 16:26:19 rrsagent, format minutes 16:26:20 I have made the request to generate https://www.w3.org/2024/01/10-voiceinteraction-minutes.html ddahl 16:26:31 rrsagent, make logs public 18:14:12 ddahl has left #voiceinteraction