RE: <device> questions

This is probably more of a proposal discussion than a requirement discussion, but I think we should be very clear about what device thing we are talking about.  There is no device tag found in http://www.w3.org/TR/html5/ and I'm skeptical that the <device> tag from whatwg has support of many as currently written.  There is the capture api that is being designed in the W3C at http://www.w3.org/TR/2010/WD-html-media-capture-20100928/ but that may not have the UI/privacy interaction we want and may not be as stream friendly.

If we do want to get involved with the Device API and Policy working group (http://www.w3.org/2009/dap/) that probably is the right way to help influence this moving forward, but we'd be quite naïve to think this is a solved issue that already meets the needs that Milan has raised.  We should probably talk about this more on this week's call when we go over the various UA/SS threads.

But it is quite possible that when we move to various proposals we can evaluate proposals that use the capture/device API along with other proposals that use different mechanisms.
From: public-xg-htmlspeech-request@w3.org [mailto:public-xg-htmlspeech-request@w3.org] On Behalf Of Satish Sampath
Sent: Wednesday, December 15, 2010 1:36 AM
To: Young, Milan
Cc: public-xg-htmlspeech@w3.org
Subject: Re: <device> questions

Yes there is a risk, though there is sufficient interest in a Device API that it will be picked up soon (I think Ericsson Labs even have a prototype implementation out now).

> Any thoughts about inviting IETF and/or Connection Peer experts to our calls?

I'm not very sure, but if we want to contribute to the Device API spec perhaps some of us should participate in their group/call rather than inviting them to ours?

Cheers
Satish

On Wed, Dec 15, 2010 at 2:24 AM, Young, Milan <Milan.Young@nuance.com<mailto:Milan.Young@nuance.com>> wrote:
Thanks Satish.  Great information.

The line between requirements and implementation is starting to blur.  Perhaps it would be worth our while to investigate the <device> approach in parallel with requirements.

The approach carries risk, because we might find <device> to be a dead end.  But then at least we would know it's a dead end and it would better frame the protocol and privacy discussions.  At present it's difficult because the potential implementation paths are so diverse.

Any thoughts about inviting IETF and/or Connection Peer experts to our calls?


________________________________
From: Satish Sampath [mailto:satish@google.com<mailto:satish@google.com>]
Sent: Tuesday, December 14, 2010 2:16 PM
To: Young, Milan
Cc: public-xg-htmlspeech@w3.org<mailto:public-xg-htmlspeech@w3.org>
Subject: Re: <device> questions

Hi Milan,

  * How does the connection peer proposal tie in with streaming speech audio?  I see support for addStream(), but this whole API seems to be oriented around peers rather than client/server.  Is this just a pattern to follow, or would we try to re-use verbatim?

Yes ConnectionPeer is currently geared towards peers and I was hoping we from this XG can influence to add client-to-server functionality as well.

  * Any thoughts on using WebSockets to transmit the data?  Lower overhead might make it a better choice for streaming compared to chunking.  Bidirectional communication would enable additional use cases and would probably simplify the process of canceling a request.

WebSockets are good if the data being sent and received is text/strings and is available to the web application. Were you thinking about the web app's script getting raw audio and sending through a websocket, or just connecting a stream from the <device> tag to a websocket? The latter seems close to the ConnectionPeer model and we may have to get in touch with the WebSockets group in IETF to discuss.

  * Is anyone aware of standards work exposing the microphone via <device>, or would this be virgin territory?  Privacy is an area where we will have a lot of requirements.

As of now we just have a generic "media" which is suitable for audio+video capture devices. I think we can bring it up in the WHATWG mailing list with our use cases. Privacy should already be an issue which <device> will be addressing and we could piggy back on that.