W3C

Timed Text Working Group Teleconference

16 January 2025

Attendees

Present
Andreas, Atsushi, Cyril, Gary, Nigel, Pierre
Regrets
Chris_Needham
Chair
Gary, Nigel
Scribe
nigel

Meeting minutes

This meeting

Nigel: Happy New Year everyone, welcome back, hope you had a good break from this.
… Today, we have on the agenda:
… * DAPT
… * IMSC 1.3
… * Charter 2025
… * AOB: "Dubtitles"
… Is there anything else anyone wants to cover or make sure we get to?

Pierre: I have an IMSC topic to cover early please

Nigel: Sure

IMSC 1.3

[WR/ARIB] Character Sets w3c/imsc#544

Pierre: [shares screen]
… ARIB kindly provides a set of characters for use in Japanese
… One of them is from ARIB STD-B62

github: w3c/imsc#544

Pierre: It's Table 5-2
… Additional symbols and characters part 1
… This table with UCS values - they're all Unicode characters I think

Nigel: Yes they're Unicode code points.

Pierre: Does Table 5-2 correspond with section 5.5?

Atsushi: No, even section 5.2 includes additional symbols in tables 5-2 and 5-3
… I may have written somewhere: originally ARIB's definition is based on JIS character set.
… So-called ARIB Gaiji was defined to include characters in tables 5-2 and 5-3 that are not in JIS.
… But at some point the ARIB definition switched from JIS encoding system to XXX
… The last bullet in 5.2 contains characters that were later included in Unicode but were "ARIB gaiji" before

Pierre: So Table 5-2 has nothing to do with section 5.5?

Atsushi: Nothing

Pierre: So from the liaison, everything is in Unicode except the last bullet which is from Table 5-2.

Atsushi: It could be possible that there was some correction to the submission from ARIB but I would not know.

Pierre: I will reference Table 5-2 explicitly, thank you.

SUMMARY: Reference Table 5-2 explicitly

Refactor the HRM w3c/imsc#586

github: w3c/imsc#586

Pierre: The current version of IMSC includes both the definition of the profile and also the HRM
… that constrains the complexity of IMSC documents.
… A conformant IMSC document is defined as one that also conforms to the HRM.
… Now the HRM is a separate document, and I think we had discussed in the past
… removing the requirement that an IMSC document conforms to the HRM
… and merely point to it as an additional specification that users might want to use.
… That has two advantages,
… a. It reduces the strong dependence between the specs
… b. it matches reality where not every IMSC application checks the HRM
… The goal is to remove the HRM from IMSC 1.3 and the suggestion is to also remove conformance
… as part of the definition of a conformant IMSC document.

Nigel: I wanted to surface this because I wanted a wider group to have visibility, than just me and Pierre.
… Also, there are other options:
… 1. Keep the hard requirement
… 2. Keep the hard requirement but flag that it is likely to be softened in the future
… 3. Change to a SHOULD
… 4. Change to a MAY

Pierre: I would not use the term MAY in practice

Andreas: As a data point, when we defined EBU-TT-D and discussed in HbbTV, and wanted to
… align with IMSC we had long discussions about the HRM and whether we include the constraint.
… We decided not to have this requirement for the HRM.
… If it is a strict requirement that leaves the possibility to have EBU-TT-D documents that do not
… conform to IMSC I don't think it is a big issue at the moment but it is likely to reflect the reality
… as Pierre is proposing.

Nigel: So you're in favour of removing the requirement and referencing as an option (i.e. 4)

Andreas: Yes

Nigel: Any other thoughts?

Pierre: I would have a different opinion if I thought this would go against practice, so there's very little risk here.

SUMMARY: Remove the hard dependency on IMSC-HRM conformance, and reference as an option for users.

Update namespace documents w3c/imsc#589

github: w3c/imsc#589

Atsushi: Sorry I haven't had time to get to this yet
… I am not sure what the process is for updating namespace documents
… I will consult the team
… The document is on github so we can do it with a pull request
… but the question is what content is required

Nigel: Would it be something different from what's in the pull request?

Atsushi: No idea actually

Nigel: OK, I suggest put a comment on the pull request when you know

Atsushi: The process says this is part of CRS publication, but we haven't done it for previous IMSC
… publications so I definitely need to do something for this.

SUMMARY: Reviews to continue

Publication

Nigel: Presumably when we merge pull requests something will get published?

Pierre: I don't know. I do have to change the README too.

Nigel: I think we need the short code reserved and then presumably echidna needs to be configured.
… I can't remember if we go all the way back to WD or straight to CR.
… There's some bureaucracy to figure out there!
… Let's take that offline.

[Pierre drops off the call]

DAPT

Nigel: I think we are just waiting for CR publication.
… Atsushi, are we waiting for anything?

Atsushi: I pinged Simone for review around year end and will do again shortly.
… In any case 2 months has elapsed since the first request so I will talk with plh soon.

Nigel: So that one HR request is the only thing holding us back?

Atsushi: i18n is finished, so yes, only waiting for Security on the Security considerations section.

Nigel: It's frustrating, can we just go ahead? When will you be able to discuss with plh?

Atsushi: Next Monday or Tuesday. I'm also frustrated!
… I'm not sure if there is anything I can do.

Cyril: A question about the security review. Why do we need another review when nothing has changed security-wise?

Atsushi: The W3C Team got a new security lead and activity formally started late last year.

Cyril: OK, we may not have had formal security review but we had a long discussion with the security team,
… that should be mentioned, right?

Atsushi: That was with the Privacy group?

Nigel: We put a fingerprint icon on, for sure

Atsushi: That should be privacy

Cyril: It's a profile of TTML2 so the security review of TTML2 applies to the profile.

Nigel: We do introduce new features that aren't in TTML2.

Cyril: Should we write down a quick paragraph to help them scope the review quickly?

Nigel: Not a bad idea

Cyril: I'll draft that

Nigel: Thank you

Cyril: Assuming we go to CR at some point, then we have to work on the implementation report.
… Nigel you did work on that already. What needs to be done?
… Do we need work, or just to get implementations to report what they do?

Nigel: Yes, we need to add the recently added features, if I didn't already do that.
… Also most importantly we need to create a test suite for the DAPT features not in TTML2.
… Then we can add into the implementation report the tests so that we can demonstrate
… the tests that each implementation passes or fails.
… Then that tells the story of how we have met the CR exit criteria

Cyril: Thinking of at-risk features, they're not new, because they're in TTML2, so they
… won't be part of the implementation report, so how will we resolve those issues?

Nigel: Interesting question.

Cyril: Implementation feedback, of course, but we will make a decision based on implementer feedback?

Nigel: Yes I think so

Cyril: We should encourage people to provide feedback on this

Nigel: I propose we add into the implementation report each at risk feature as a row,
… and then show which implementations implement each feature so we can make a clearer decision,
… even though there's no formal requirement for DAPT-specific tests with them.

Cyril: I think it's a good idea

Cyril: Hard to tell now, but what is your expectation in terms of timeline?
… Can we publish Rec in 2025.

Nigel: I expect so, yes.
… Not sure if I mentioned it yet, but the EBU Eurovox project has already implemented a DAPT output.

Charter 2025

Nigel: Not much to say except that there has been a notification of the new charter work,
… based on what we did at the end of last year. Thank you Atsushi.

Atsushi: We are now at HR stage.
… Advanced notice was sent to AC Reps, external parties like public-new-work mailing list of W3C.
… HR should finish within 2 months and we will go to AC review right after that.
… Our changes are not large in terms of text, I don't expect to get large amounts of comment from the HR
… groups or AC.

Nigel: In other words, we can forget about this for a while!

Atsushi: Yeah

Nigel: The only thing is there are some TODOs on it, for example when we've published IMSC 1.3
… in some form then we need to update the links to it.

Atsushi: For that point, to publish FPWD the spec needs to be in the Charter,
… so I think we should publish the ED for that.

Nigel: I don't agree, I think it is already in scope of the current charter to produce new versions
… of existing Recs.

Atsushi: Could be part of Profile of TTML

Nigel: No doubt, it definitely is!
… If the next stage is FPWD then that's what I'd like us to do.
… Anything else on Charter?

nothing else for now

Nigel: I will remove this from the ongoing agenda unless you want to raise it again, or something comes up.

AOB - Dubtitles

Gary: Background: "Dubtitles" are what people have been calling subtitles that use the dubbing script
… for the text. Normally translation subtitles have a different timing match to the original audio.
… With dubtitles the timing matches the dubbing audio.
… For viewers, sometimes there's a lot of mismatch so it can be a frustrating experience.
… You might hear something in the original audio but then the translation cue comes up later or earlier.
… My question is: is there something we can do technology-wise to help reduce the use of dubtitles.
… One of the issues is likely the cost, where if you only generate one thing then it's cheaper than
… generating a subtitle that matches the original audio and then a separate one that matches the dubbing script.
… I don't know if there's something in DASH or HLS or somewhere down the stack that's missing that makes
… it less likely that people will create both subtitle tracks.

Cyril: I'm not sure I understand why you want to reduce the use of dubtitles?

Gary: If I'm watching in the original audio then I want subtitles that match the timing of the audio.

Cyril: "captions" in the US

Gary: Yes. But what's happening is that the timing matches the dubbed audio.

Cyril: In what language?

Gary: This comes up a lot in anime, where if I try to watch in the Japanese original audio but the captions
… are the English dubbed audio captions.

Cyril: The problem is the subtitles for English are created from the dub instead of the transcribed Japanese?

Gary: Yes, the main issue is the timing of when the cues are shown, which are derived from the dubbed audio.
… If you're watching in the original Japanese the translation cues don't show up when you're expecting
… based on the Japanese audio.
… The specific content of the subtitles is less of an important issue.
… The dubbing translation is usually good enough.

Cyril: That's related to DAPT, where the use case was to avoid discrepancies between the dubbed audio
… and the translated subtitles in the same language.

Gary: I think it's related, but yes there's weird discrepancies.

Andreas: From my understanding this needs a linkage between the audio language and the
… translated subtitles, and I don't think that exists.
… You can watch the same video image with any of the audio or subtitle languages offered.
… Right now there's no solution.
… You could of course use those dubtitles with the dubbed version, which should fit.

Gary: That should be fine because the timing should align in that case.

Cyril: It's also a constraint you can put in the workflows, not to create dubs with timing that is too different
… to the original.

Gary: It's more complicated because in the dub sometimes there are extra lines spoken where there's no
… dialogue in the original. Then you see extra subtitles when nobody is speaking.

Cyril: Sounds like an annotation to show that an event is not in the original language.
… In DAPT you would have... er, not sure how we support that.

Nigel: I've got beads of sweat worrying about if we support this!

Cyril: It's like audio description, describing something not being spoken.

Gary: They are spoken, just only in the dubbed audio.

Cyril: It's an interesting use case. Can we work on a concrete example?
… I would be interested to see content that demonstrates this.
… Annotating extra subtitles could be done, and the player could act on that.

Andreas: Would a simple solution be to distribute the metadata about the source language,
… so if you combine it with some other audio track then there could be problems.
… It's not just a timing problem, but also about the content, because subtitles from the dubbed version
… used for the original do not match. That's another possible issue.

Gary: Yes, if you know the original language and watch the subtitles in that language it's an issue.
… For the anime example it's not an issue for me, Japanese audio and English subtitles,
… but for dubtitles when the timing matches up it's not an issue.
… When the timing is so different then it makes it hard to watch.
… I agree maybe the issue is, in downstream like DASH and HLS, we need a way to associate specific
… subtitle tracks with specific audio tracks.
… With DAPT it could be useful if there are these annotations already.

Nigel: I feel that in DAPT we have enough metadata to add the annotations but it needs checking.

Meeting close

Nigel: Let's adjourn a little over time today, thanks everyone.
… Next call in 2 weeks.
… [adjourns meeting]

Minutes manually created (not a transcript), formatted by scribe.perl version 242 (Fri Dec 20 18:32:17 2024 UTC).