14:17:55 <RRSAgent> RRSAgent has joined #voiceinteraction
14:17:59 <RRSAgent> logging to https://www.w3.org/2024/04/24-voiceinteraction-irc
14:18:21 <ddahl> meeting: Voice Interaction
14:18:28 <ddahl> scribe:ddahl
14:18:35 <ddahl> chair:debbie
14:18:45 <ddahl> topic: reference implementation
14:24:22 <ddahl> rrsagent, make logs public
14:24:32 <ddahl> rrsagent, format minutes
14:24:33 <RRSAgent> I have made the request to generate https://www.w3.org/2024/04/24-voiceinteraction-minutes.html ddahl
14:26:18 <ddahl> https://github.com/w3c/voiceinteraction/tree/master/source/w3cipa
14:27:13 <ddahl> agenda: https://lists.w3.org/Archives/Public/public-voiceinteraction/2024Apr/0010.html
14:37:14 <ddahl> present: dirk,gerard, debbie
14:39:17 <ddahl> dirk: review reference implementation
14:39:29 <ddahl> ...ChatGPT and Mistral
14:40:38 <ddahl> ....framework is mainly headers
14:41:16 <ddahl> ...component that accesses GPT, also demo program
14:42:21 <ddahl> dirk reviews SOURCE.md
14:43:32 <ddahl> input listener for modality inputs
14:43:54 <ddahl> ... in this case just selects the first one
14:44:09 <ddahl> ...goes to ModalityManager
14:44:28 <ddahl> ...can add modality components as you like
14:45:30 <ddahl> ...startInput and handleOutput
14:46:22 <ddahl> ... this is part of the framework, so Royalty Free
14:48:24 <ddahl> ...modality type is a free string so it can be extensible
14:48:46 <ddahl> ...only text is implemented in the reference implementation
14:50:39 <ddahl> ...some modality components could be both input and output
14:54:08 <ddahl> ...one instance that knows all listeners and that all modality components would know
14:55:15 <ddahl> ...looking at one example of a modality component, textModality
14:59:16 <ddahl> debbie: can there be more than one InputModalityComponent?
14:59:21 <ddahl> dirk: in theory, yes
15:02:34 <ddahl> ...we might have scaling issues with multiple text inputs, for example
15:03:36 <ddahl> rrsagent, format minutes
15:03:37 <RRSAgent> I have made the request to generate https://www.w3.org/2024/04/24-voiceinteraction-minutes.html ddahl
15:07:42 <ddahl> debbie: take "first" out of name "TakeFirstInputModalityComponent" to make it more general
15:08:24 <ddahl> dirk: moving on to DialogLayer, IPA Service
15:09:18 <ddahl> ...IPA for both local and anything else we have
15:10:17 <ddahl> ....ReferenceIPAService consumes data from Client
15:11:31 <ddahl> ...could serve multiple clients or if we have local and other IPA services
15:11:45 <ddahl> ...no DialogManager in place
15:13:49 <ddahl> ...if there was one, the IPA service would send the input to it and then after that the IPA service would forward the output back to the client
15:14:24 <ddahl> ...the ExternalIPA/Provider Selection Service
15:14:58 <ddahl> ...the Provider Selection Service for now only knows about ChatGPT
15:19:37 <ddahl> ...IPA provider supports input from different modalities
15:20:48 <ddahl> debbie: should we standardize on define modality types, e.g. "voice" vs "speech"
15:21:55 <ddahl> dirk: would like to talk about ProviderSelectionStrategy and how components are glued together
15:22:15 <ddahl> debbie: we can talk more in the next call
15:25:09 <ddahl> ...could we list the parts of the architecture that aren't implemented yet?
15:25:09 <ddahl> dirk. that might make sense
15:26:20 <ddahl> debbie: could there be an UML diagram?
15:26:20 <ddahl> dirk: there could be more diagrams
15:26:46 <ddahl> ...could link from code to specification
15:27:42 <ddahl> dirk: next time talk about the provider selection strategy and how to chain everything together
15:27:47 <ddahl> debbie: will try runnin
15:27:57 <ddahl> s/runnin/running
15:28:36 <ddahl> dirk: demo running with ChatGPT
15:30:28 <ddahl> gerard: which version of Mixtral do you use?
15:30:58 <ddahl> ...open source version
15:31:20 <ddahl> hugues: the next version will not be open source
15:31:41 <ddahl> gerard: the approach is mixture of experts
15:32:27 <ddahl> dirk: what happens if we ask both at the same time?
15:32:50 <ddahl> ...would receive them both
15:33:05 <ddahl> gerard: could use an LLM to summarize
15:33:25 <ddahl> ...that's what Mixtral is using with the Mixture of Experts
15:34:34 <ddahl> rrsagent, format minutes
15:34:35 <RRSAgent> I have made the request to generate https://www.w3.org/2024/04/24-voiceinteraction-minutes.html ddahl
17:34:20 <ddahl> ddahl has left #voiceinteraction