See also: IRC log
<burn> trackbot, start telcon
<trackbot> Date: 16 December 2010
<burn> Scribe: Robert_Brown
<burn> ScribeNick: Robert
Dan: (no comments) last week's minutes approved
Dan: no comments
michael: not much mail on this, Bjorn agreed in mail, no other mail comments. seems reasonable
<mbodell_> proposed req: Web application must be able to encrypt communications to remote speech service
Dan: asked for objections, no objections voiced
Milan: not sure we're aligned on
the emphasis behind this requirement. maybe should put it on
hold. some people are prioritising schedule ahead of
... put it on hold and see how the other issues we discuss this week play out
Bjorn: has anybody had experience where this sort of requirement is needed? it seems redundant
<bringert> I got disconnected
Dan: sometimes to prevent avoiding certain architectures
Milan: intended to avoid the sessions/sockets issue. but lets get on dissing the other topics and get back to this one
Bjorn: i wouldn't consider it high priority, but okay keeping it for now
Dan: this is certainly in scope
Bjorn: it's already possible and doesn't need a new requirement. just use an xmlhttp request.
Dan: there may be some benefit to having a unified approach
Bjorn: agreed there's a benefit but not high priority
Dan: looks like we have consensus on keeping it
<mbodell_> proposed req: Web applications must be able to request NL interpretation based only on text input (no audio sent).
Michael: a fair bit of discussion in mail, but it seems people are okay keeping this
Bjorn: okay to have as a requirement, lower priority, if I was making the proposal I wouldn't add it because of the added complexity
<mbodell_> proposed req: Web applications must be able to request recognition based on previously sent audio.
Michael: no objections? [resounding silence...]
Michael: discussion on whether we need it and whether cookies support it?
Milan: not thrilled, but okay to
call this one good enough
... cookie gets 90% of use cases
Bjorn: do you want to add a requirement like existing mechanisms should be used to manage sessions or something like that
Milan: how about the way it's worded now?
Bjorn: text in original email is okay with me
Olli: okay with me too
<Milan> Robert nervous about defintition of word session
<burn> robert: wants to confirm meaning of "session". different from what we do in web apps?
<burn> robert: is there any use case?
<burn> bjorn: yes. could consider a speech API that does not pass on cookies that are set
<burn> milan: e.g. a native agent proposal. user agent would be required to tack on cookies
<burn> robert: can live with this. details will become apparent with the proposals
<burn> bjorn: IETF specs use the notion of "stateful session" when discussing cookies
<mbodell_> proposed req: Web application and speech services must have a means of binding session information to communications.
michael: sounds like we have consensus
Bjorn: okay with Milan's restatement in mail
Michael: concerned that this breaks our privacy requirements
Milan: but that's broken (paraphrase)
Michael: if I'm the only one who's nerveous I'm okay taking Milan's text
Bjorn: if those mechanisms don't satisfy privacy requirements, we can look at improving them.
Marc: is it part of our specification to make a position on who does it?
Bjorn: xmlhttp talks about web app but implies UA requirements
Dan: nerveous but won't object. in prioritisation we may need to be more precise
<mbodell_> proposed change: fpr30 becomes Web applications must be allowed at least one form of communication with a particular speech service that is supported in all UAs.
<marc> my question was about confirming that at this stage we are not taking any decision how the communication between the web app and the speech service is realised, whether the UA plays a standardised role or not.
Dan: agreed, move on
<marc> confirmed that this decision is *not* taken at this stage.
<marc> the new requirement is better because it makes this less explicit.
Bjorn: besides efficiency, are there any reasons to add the requirement?
Michael: existing requirements relate to this (barge-in)
Milan: it's efficiency. but if you were going to do real barge-in in most of your transactions, it would be an issue
Bjorn: if the client wants to stop sending audio, it can send a marker saying it's done
Milan: that's what I'm asking for
Bjorn: sender cancelling is easy with HTTP. receiver cancelling is difficult
Milan: how would end of speech be indicated
Bjorn: some sort of end-of-audio
packet, which handles the sender cancelling
... why do we need this?
Milan: the user agent may not be able to detect when done
Bjorn: would server or client do that?
Milan: the client
Bjorn: should split into two
discussions: 1 client aborting recognition (fine and required
and trivial); 2 client aborting synthesis
... implied by FPR17
Michael: that says the user can abort it
Bjorn: need a separate requirement that web application should be able to cancel audio capture
Marc: we used the term "abort" intentionally, with privacy concerns in mind
Bjorn: duplicate FPR17, replacing user with web app
<mbodell_> proposed new req: While capture is happening, there must be a way for the web application to abort the capture and recognition process.
Bjorn: fine with what Michael
... [no other objections] lets move on to synthesis
... client wants to abort playing of long synthesized speech. if there's no way for the client to signal the server, the only option is to tear down the connection
... this may have latency implications to establish a new connection
Milan: there's a lot of work that
goes into establishing a TCP socket. Email triage is a good
example. App reads a few sentences of a message then the user
... it would be awkward if the mail app just read the first sentence
Bjorn: or the app could read a sentence at a time until it decides to move to the next message
Milan: not asking for interruption (existing requirement), but to cancel it all the way to the server
Bjorn: reluctant to add a
requirement of going all the way to the server
... propose "web application must be able to abort TTS output"
Milan: but Bjorn has already to do this for reco, why not TTS?
Bjorn: reco is required, and the sender aborts by sending up a token. this is different, because the receiver is aborting
Milan: but with reco, the server is sending back ack's while the client is speaking, so there is a bi-directional mechanism
Bjorn: are you saying a bidirectional communication is already required?
Milan: we have the requirement that speech has begun and streaming
Bjorn: speech detection is done on the client
Milan: nerveous about detection
in the client
... FPR21 apps should be notified when capture starts
... until we have reco, we can't say that speech has begun, and we can't do hotword from the client
Bjorn: notify -that- speech has begun, not -when- it has begun
Milan: this is part of the problem of not having detailed descriptions on this. I brought this up back in the F2F meeting, but didn't catch the nuance of the word "that"
Bjorn: no assumption that detection runs on the client, but also no exclusion of this
Milan: but if it runs on the
server, then you need bi-direction communication
... and if so, it doesn't seem to be a stretch to say we need this for synthesis
Bjorn: i agree with the analysis, but probably wouldn't propose an API for this
Michael: we shoudl agree on whether or not it's a requirement, then prioritise in the next stage
<mbodell_> proposed req: Web application must be able to programatically abort tts output.
Bjorn: can we agree that it's a requirement for the web app to abort TTS, without any specific requirement on how thsi affects the server
Milan: sounds fine
Michael: (silence) sounds like we have consensus
Bjorn: so the other requirement is that when the client aborts TTS, it should not need to tear down the connection
Marc: is this about functionality or efficiency? if it's about efficiency, the discussion should occur later, when we discuss implementation
Milan: but it's so fundamental it would be crippling not to have this
Bjorn: how about "aborting TTS should be efficient"?
<mbodell_> proposed req: Aborting the TTS output should be efficient.
Michael: sounds like we have consensus
Bjorn: "TTS output" rather than
... one is the effect on the user experience, the other is the effect on efficiency
Michael: is there a set of requirements out of that discussion?
Bjorn: no it's a proposal
Milan: it shows a lot of promise and if we started early we could get done sooner
Bjorn: there's some serious politics going on there
Michael: WHATWG doesn't really represent all browser manufacturers
Milan: could the audio working group handle this?
Michael: they're more about
mixing and analysis, rather than capture
... IE wouldn't tackle this area until it's under some w3c group
Milan: it would be in our group's interest to get some sort of audio capture API into HTML
oops, that should have been Bjorn
Michael: UI is geared around web cam capture
Milan: people have been working on audio capture since 2005, and we only started this year
Michael: but the use cases are different
Bjorn: is there an audio chat
... could we specify an API required for speech without it being general purpose?
Michael: we should propose what we need and explain why we need it
Bjorn: if we don't have a general API for app-specified network recognition, we can still have reco with the default recognizer
Olli: would it be easiest to co-author it with the whatwg and then propose that the HTML wg pick it up
Bjorn: that's my preference
Marc: if the browser captured audio according to ther requirements for speech recognition, then we wouldn't need any specific device API
Michael: an alternative is to finish discussing requirements, then look at proposals, for which there may be a spectrum of approaches
Bjorn: there's no reason to exclude a particular approach at this point
Milan: concerned that device API has a promise and if we don't work together it won't happen
Marc: we're expected to look at the pros and cons of various options and maybe make a decision, or if not, at least recommend options
Dan: people can propose more
requirements later on, but we should move on to
... begin prioritization in January, but between now and then, review the requirements and talk about those you don't feel are clear enough for you to prioritize
Michael: please send description text where you think it's missing
Milan: would prefer that the chairs propose a description and participants riff on that
Dan: prioritization is a function
that will naturally work out issues at the next level of
... So the first thing people should do is review the requirements, and if you can't prioritize, start a conversation
Michael: I will send out another update soon, and you'll have a couple of week to review as Dan suggests
Milan: it'll be chaos. 50 requirements. 6 groups here
Dan: if this turns out to not
work, we'll change strategies
... but I think we'll probably have a very small number of threads
... Plan to have calls at the same timeslot in January, in case we need them
Marc: Michael, could you restructure the list of requirements by topic?
Michael: will move section 3 to
an appendix, and can potentially reorder section 4. I'll make
... I'll see what factors out
Great work everybody!
This is scribe.perl Revision: 1.135 of Date: 2009/03/02 03:52:20 Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: RRSAgent_Text_Format (score 1.00) Succeeded: s/...:/... / Succeeded: s/obvious // Succeeded: s/an way/a way/ Succeeded: s/the synthesis/the TTS output/ Found Scribe: Robert_Brown Found ScribeNick: Robert Default Present: Michael_Bodell, Olli_Pettay, Milan_Young, Bjorn_Bringert, Dan_Burnett, Debbie_Dahl, Robert_Brown, Marc_Schroeder, +1.732.507.aabb Present: Michael_Bodell Olli_Pettay Milan_Young Bjorn_Bringert Dan_Burnett Debbie_Dahl Robert_Brown Marc_Schroeder +1.732.507.aabb Agenda: http://lists.w3.org/Archives/Public/public-xg-htmlspeech/2010Dec/0144.html Found Date: 16 Dec 2010 Guessing minutes URL: http://www.w3.org/2010/12/16-htmlspeech-minutes.html People with action items:[End of scribe.perl diagnostic output]