HTML Media Task Force Teleconference -- 19 Mar 2013

<trackbot> Date: 19 March 2013

<paulc> Good morning, Adrian (evening in Beijing)

<jon> zakim aabb is me

<adrianba> ScribeNick: adrianba

<scribe> Scribe: Adrian Bateman

Roll call, introductions and selection of scribe

paulc: this is a call of the Media Task Force of the HTML WG
... focus on Media Source today - alternate every two weeks
... ask for people to identify themselves
... objective to get through liaison questions and first bugs
... to give them feedback on bugs
... hope editors ready for that - know there have been some responses on individual bugs
... I'm Paul Cotton, from Microsoft, co-chair

adrianba: Adrian Bateman, Microsoft, Internet Explorer team

<Michael_Thornburgh> i'm from Adobe

[... members of the call identifying themselves]

<jon> I'm Jon Piesing from TP Vision, what used to be the TV business of Philips. I'm the chair of the browser group in OIPF and of the specification group in HbbTV.

<SimonW> Simon Waller from Samsung, I work on IP standardisation for TVs in OIPF and HbbTV

<Steve_Morris> I'm Steve Morris from Espial, working in both HbbTV and OIPF

<giuseppep> Giuseppe from Opera, working on Opera TV product. co-chair of the web&tv and also active in OIPF and HbbTV

<stevep> Stephen Perrott from the BBC in the UK. Here as part of the OIPF and DTG guest contingent.

paulc: that's done the introductions

<jbHenry> I'm J.B. Henry, from Orange group, working with HbbTV (and also OIPF)

MSE Liaison questions and new bugs

<acolwell> Aaron Colwell from Google. I work on Chrome HTML5 video and I'm editor of the Media Source Extension spec.

<David_Corvoysier> I'm from Orange, both attending W3C HTML WG & Web &TV IG and OIPF

<guy> Guy Paskar, Peer5, Peer assisted delivery network utilizing MSE for creating the video from multiple sources.

paulc: first topic is the question from the liaison groups

<stephane> Stephane Lejeune and Alain Tholon with NDS in France working on HbbTV.

MSE questions from Open IPTV Forum, HbbTV and the UK DTG

<BobLund> Bob Lund from CableLabs working on browsers for TV, and DLNA and W3C related efforts.

<nsm> Nigel Moore - Sony Europe. Working on IPTV and web standards for CE devices in DTG, OIPF and HbbTV groups.

http://lists.w3.org/Archives/Public/public-html-media/2013Mar/0039.html

<ddorwin> David Dorwin from Google Chrome; co-editor of Encrypted Media Extensions.

<paulc> Response: http://lists.w3.org/Archives/Public/public-html-media/2013Mar/0048.html

<skaiser> Stefan Kaiser from Fraunhofer FOKUS, working on adaptive streaming / DASH applications

paulc: Q1. Are there plans for test cases for MSE?

acolwell: i was under the impression that the process requires this

paulc: process in W3C is specs go through series of Working Drafts
... First WD has been published, typically publish new WD every 3 months
... when the WG is done, it approves going into Last Call, Last Call Working Draft, and then look for public review
... once all comments are processed, spec moves to Candidate Recommendation
... that transition requires WG approval and approval of W3C Director
... at this stage there is a public call for implementations
... usually need test cases to show interop and that spec is clear enough for third party to implement from spec
... not unusual for there to be test cases before CR
... only at CR is mandatory
... is there a reason for the question?

jon: for the three organisations a test suite is important
... we were not that familiar with w3c practice
... the only immediate question is one of timescales
... think we're done with this question

SimonW: i think this will be covered with schedule

paulc: Q2. Do you believe MSE contains the features needed for a DASH player that fully supports the DASH profiles defined by MPEG and by DASH-IF? If a DASH player on top of MSE could only support a subset of these profiles, is there any information on what could not be supported?

markw: for iso bmff i'm fairly sure MSE can support a DASH player
... haven't done this for MPEG2-TS

acolwell: also my assessment

BobLund: we provided bitstream definition for MPEG2-TS and this was driven by requirements for DASH support

adrianba: appreciate specific bug reports if people think something is missing or broken

Q3. Has any thought gone into how MSE could be used with a media element whose playback rate was other than 0 or 1? Particularly rates greater than 1 or less than 0.

acolwell: two ways this could be handled
... this should be signaled by setting rate on media element just like the non MSE case
... either just append faster because media engine needs faster
... or if you could use difference sourcebuffers for more sophisticated output (e.g. key frame) and switch tracks with html5

SimonW: what happens to the timelines if you switch sourcebuffer and then you switch back
... you want to switch into a different time

acolwell: my assumption, if you're using a different buffer you're still using the real timeline
... so it's really just track selection
... the current playback position is just moved forward normally
... depending which sourcebuffer you're using you might only be decoding keyframes
... but they're all the same timeline

<adrianba_> ScribeNick: adrianba_

paulc: Q4: Is there a complete description for the handling of encrypted content with an embedded DRM?

acolwell: this is covered by EME
... there was care taken to make sure encryption cases were handled but not called out

<paulc> EME = Encrypted Media Extensions

acolwell: because some people want to support MSE but not necessarily encryption

See http://lists.w3.org/Archives/Public/public-html-media/2013Mar/0048.html for example

<markw> yep, that's correct

johnsim: same for ISO

jon: that is the answer i was expecting

<adrianba> ScribeNick: adrianba

paulc: we've kept MSE and EME separate but there are people planning to implement both
... Q5. The broadcasters in our 3 organisations have a requirement to know when a particular advert starts being played and stops being played so that they can report this information back to an advertising server. It seems to us that the (only?) way to achieve this with with MSE is to remember a) where in the timeline the advert has been inserted (as defined by timestampOffset) and b) the length of the advert. As the content is played, the app wo

uld poll the currentTime property of the media element and test if this was within an advert. Is this how you ... would expect this requirement to be met with MSE?

acolwell: this is one way to do it
... i would expect to use the existing html5 text track to do this
... create text cues with the metadata
... for the start and end
... when the playback reaches those cues it will fire an event
... without the need for polling
... could use inband tracks for this too

BobLund: here at cablelabs we've implemented both approaches and would be happy to provide details to anyone interested

paulc: thanks for the offer

MSE Liaison bugs

paulc: may have been more bugs filed - there were 5 when i created the agenda

<paulc> https://www.w3.org/Bugs/Public/show_bug.cgi?id=21298

<BobLund> Contact b.lund@cablelabs.com for implementation details of using text tracks for advertisement audits

paulc: each bug has a history in it

Bug 21298 - making it easier for people to understand MSE

https://www.w3.org/Bugs/Public/show_bug.cgi?id=21298

acolwell: the actionable piece here was about creating a diagram
... about how appended data flows through the system
... open to more suggestions but this is for implementers so a lot of prose makes it difficult to keep clear what is normative
... could create a separate document for content owners

SimonW: helpful to have a high level overview for how it all works together
... a separate document perhaps
... for example implementation guidelines for apps

paulc: not uncommon to produce a primer in W3C
... for example XML Schema
... Part 0 is the primer and read more often than the normative spec
... that model or introductary material in an annexe also possible

jon: if you want good feedback need to reduce the learning curve

Bug 21299 - memory management in MSE

<paulc> https://www.w3.org/Bugs/Public/show_bug.cgi?id=21299

https://www.w3.org/Bugs/Public/show_bug.cgi?id=21299

acolwell: this bug is around what assumptions MSE is making about how memory is managed
... easiest way to think about this is SourceBuffers contain trackbuffers
... which hold timeranges
... remove() allows removing ranges from the timeline
... only assumption is that you have these buffers of coded frames and you can remove time ranges
... and encoding dependencies are taken into account
... don't believe any other assumptions about algorithms or methods of storage

jon: this section was added after our F2F
... two immediate things spring to mind
... does the app have to call remove()

acolwell: not required any more for remove to be called
... we did have this requirement for a while but went back to the original model
... intent is that if you don't call remove then UA will manage buffer for you
... and attempt to ensure playback doesn't stall
... if app wants different algorithm then it can use remove
... also heard that some apps have contractual obligations about how much can be buffered

jon: that alleviates one concern - would apps know enough to meaningfully call remove
... but if use case is limit for rights reason then that's fine

acolwell: if app happens to know something, for example if user executes a seek and knows won't need old data then app could remove
... this might be info UA doesn't have

Bug 21300 - lack of clarity around appendStream

https://www.w3.org/Bugs/Public/show_bug.cgi?id=21300

acolwell: MSE spec doesn't talk about what creates Stream object
... MSE doesn't care but common use case is XHR
... XHR is XMLHttpRequest object
... primarily how apps that use MSE will fetch media data from the media server
... using JavaScript to initiate HTTP requests

Bug 21301 - timelines in MSE and HTML5

https://www.w3.org/Bugs/Public/show_bug.cgi?id=21301

acolwell: if we don't have questions, how do we close bugs?

paulc: want to answer these at the end
... but if editors think they've answered the question, they resolve the bugs

jon: the other option would be for the chairs to have an offline discussion

bug 21301

acolwell: this was a clarification of terms - spec uses presentation timeline and timeline without being defined
... so resolution is to use 'media timeline' which HTML5 uses
... trivial change

Bug 21302 - Support for inserting adverts as non-fragmented MP4 files

https://www.w3.org/Bugs/Public/show_bug.cgi?id=21302

acolwell: my suggestion is not to support this but i'm not sure how to attack this bug
... non-fragmented files do not follow the model of everything else in the spec
... and while i understand current ad systems do not support fragmented files i think they will
... adding this might cause legacy helpful in the short term but not the long term

jdsmith: while it is technically possible to make this work it is definitely a problem
... [some of the problems listed]
... our position is to not spend time on this

acolwell: seems like a business problem not a technical problem

jdsmith: tools are available to do this processing

jon: the reason you didn't get a response is i don't think you have the advertising-funded broadcasters here
... need to get them to think about this

<jon> Correction: The people who aren't here are the advertising-funded broadcasters.

acolwell: if it's not clear already i think you'll face strong pushback on supporting non-fragmented MP4

Bug 21333 - consequences of requirement in section 8 on initialisation segments

https://www.w3.org/Bugs/Public/show_bug.cgi?id=21333

jon: typical example is you might have a content item where there is one video track and the audio track is multi-channel
... video might be the same codec but audio is different
... other example is content has multiple audio tracks but ad doesn't
... this seemed to force us into different SourceBuffers
... what user experience would you get at the change between SourceBuffers

adrianba: [summarises comments in the bug]

acolwell: channel count can be supported in a single sourcebuffer but codec change, as adrian says, for implementation reasons you don't usually support codec changes mid-presentation, especially for hardware decoding

<markw> I was going to agree this is a Media Element problem: what happens when you play a normal media file in which one audio track stops and another starts at some point in time ? Does the Media Element automatically switch tracks ?

acolwell: so using two sourcebuffers to handle this would mean there is a track change
... track changes are not considered seamless at this time

jon: this needs more discussion
... if an audio track disappears that would seem to require a sourcebuffer change
... we want something as seamless as possible
... for ad insertion

acolwell: that's not because of the spec - the content you're using forces this
... because you're using content that can't be guaranteed to be seamless

<jon> There's at least one more bug to come.

paulc: chairs are going to have to continue our discussions
... ask editors to continue to close bugs as appropriate and keep discussing others
... liaison guests - please let us know which need more clarifications
... also would you like another meeting after a subsequent meeting on your side
... mark as NEEDSINFO or whatever is appropriate
... i'll work with liaison chairs to make sure process is understood

acolwell: should some questions that are clarification be mails instead of bugs
... bugs are for things that need to be changed
... mails for discussion

paulc: yes, if there are questions that should be answered by reading spec, can start in the list

<giuseppep> can non HTML WG members give contributions to the group?

paulc: tend to encourage to use bugs when can point to specific point in the spec and have a proposal for changes
... will provide minutes to our partners

jon: like to express thanks for time and energy put into answering questions

Adjournment

paulc: adjourned, thanks

<jon> ?

<jon> quit

- DRAFT -

HTML Media Task Force Teleconference

19 Mar 2013

Attendees

Contents

Roll call, introductions and selection of scribe

MSE Liaison questions and new bugs

MSE Liaison bugs

Adjournment

Summary of Action Items

Scribe.perl diagnostic output