Re: [TTML-WEBVTT] How to map multiple <p> that share same time range to WebVTT

Hi Boy,

> - Does that mean you're of the opinion that the specification should be changed in some way?

Probably -- or the specification should be deprecated.

Fundamentally, the subset of TTML [1] supported by the specification
does not correspond to common subsets, e.g. IMSC, and supports only a
limited subset of TTML features, e.g. see #timing constraints.

The specification has not been touched in 6 years.

@Andreas Tai who contributed to both the WebVTT-TTML mapping
specification and ttconv might have an opinion.

>  there seems to be no reference to ISDs in the document, so I'm not sure how a person reading this specification
> should know that a conversion to ISDs needs to take place first, or that ISDs play some other role in the conversion process.

I do not disagree.

Best,

-- Pierre

[1] https://w3c.github.io/ttml-webvtt-mapping/#the-ttml-to-webvtt-profile

On Fri, Nov 12, 2021 at 4:43 AM Boy van Dijk <boy@unified-streaming.com> wrote:
>
> Hi Pierre,
>
> Thanks for your response and sorry for leaving it waiting a little bit. I anticipated there would perhaps be more opinions on this.
>
> Unfortunately, I believe my initial message might not have been entirely clear because the original formatting was removed. For example:
>
> "Every <p> is mapped to a WebVTT cue."
>
> Is not something I wrote myself like it might have seemed, but a direct quote from the TTML WebVTT mapping spec (https://w3c.github.io/ttml-webvtt-mapping/).
>
> As you say this quote might not represent the right strategy:
>
> - Does that mean you're of the opinion that the specification should be changed in some way?
> - Or are you saying I'm somehow missing something relevant in the specification (which might very well be these case!)?
>
> As for the second option, and the strategy you propose, using ISDs: there seems to be no reference to ISDs in the document, so I'm not sure how a person reading this specification should know that a conversion to ISDs needs to take place first, or that ISDs play some other role in the conversion process. Although working with ISDs might very well be the better approach of course, as you indicate.
>
> Please let me know your thoughts.
>
> Regards,
> Boy
>
> On 4 Nov 2021 at 01:23:48, Pierre-Anthony Lemieux <pal@sandflow.com> wrote:
>>
>> Hi Boy,
>>
>> Every <p> is mapped to a WebVTT cue.
>>
>>
>> I am not convinced that this is the right strategy.
>>
>> I would instead map each intermediate synchronic document (ISD) to a
>> WebVTT cue, so that no two cues have overlapping temporal ranges.
>>
>> This is the approach taken by https://github.com/sandflow/ttconv.
>>
>> Best,
>>
>> -- Pierre
>>
>>
>>
>> On Wed, Nov 3, 2021 at 4:16 PM Boy van Dijk <boy@unified-streaming.com> wrote:
>>
>>
>> Hi,
>>
>>
>> I represent Unified Streaming and I'm seeking your expertise about mapping to WebVTT the following bit of TTML:
>>
>>
>> <div begin="00:00:46.320" end="00:00:48.360">
>>
>>     <p style="singleHeightStyle" tts:textAlign="center" region="region-20">
>>
>>         <span xml:space="preserve">&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;</span>
>>
>>         <span xml:space="preserve" tts:backgroundColor="black">&#xA0;You&#xA0;guys&#xA0;wanna&#xA0;story?&#xA0;</span>
>>
>>         <span xml:space="preserve">&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;</span>
>>
>>     </p>
>>
>>     <p style="singleHeightStyle" tts:display="inlineBlock" xml:space="preserve" region="region-20">&#xA0;</p>
>>
>>     <p style="singleHeightStyle" tts:textAlign="center" region="region-20">
>>
>>         <span xml:space="preserve">&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;</span>
>>
>>         <span xml:space="preserve" tts:backgroundColor="black">&#xA0;(MEN&#xA0;CHEERING)&#xA0;</span>
>>
>>         <span xml:space="preserve">&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;&#xA0;</span>
>>
>>     </p>
>>
>> </div>
>>
>>
>>
>> Or, my problem in simplified form:
>>
>>
>> <p begin="00:00:46.320" end="00:00:48.360">
>>
>> You guys wanna story?
>>
>> </p>
>>
>> <p begin="00:00:46.320" end="00:00:48.360">
>>
>> (MEN CHEERING)
>>
>> </p>
>>
>>
>>
>> From what I understand of the spec, what needs to happen is pretty easy, because:
>>
>>
>> Every <p> is mapped to a WebVTT cue.
>>
>>
>>
>> Which would result in:
>>
>>
>> 00:00:46.320 --> 00:00:48.360
>>
>> You guys wanna story?
>>
>>
>> 00:00:46.320 --> 00:00:48.360
>>
>> (MEN CHEERING)
>>
>>
>>
>> Considering that these two cues have the exact same start and end time, does their sequence carry meaning? I'm not sure, but I believe this result is far from ideal, especially after applying the very last step listed in the steps to convert TVTT to WebVTT:
>>
>>
>> The last step is to sort these cues from earliest to latest time, based on each cue's beginning timestamp.
>>
>>
>>
>> The result of which will be that either one of the cues listed is first with the other listed second. Randomness, it seems.
>>
>>
>> Okay, long email you might think but does this actually have any practical implications? Yes! If you play this back in a recent native HLS player on an Apple device you get completely unusable results (not all cues are presented and the one that are, aren't necessarily presented at the right time either).
>>
>>
>> So, I believe my questions are the following:
>>
>>
>> Is my understanding of the mapping spec as I presented it above correct?
>>
>> If my understanding is correct, does my example simply represent an edge case that isn't properly covered by the spec, or is only Apple's WebVTT parser to blame here?
>>
>> If this is an edge case not covered by the spec, what would be the way forward?
>>
>>
>>
>> Happy to hear your input and thanks you for your thoughts.
>>
>>
>> For those interested, I created a simple Tears of Steel-based test stream without audio: https://origin.unified-streaming.com/public/tkt32756/main.m3u8
>>
>>
>> It contains the following WebVTT:
>>
>>
>> WEBVTT
>>
>>
>> 00:00:05.000 --> 00:00:05.000
>>
>> You guys wanna story?
>>
>>
>> 00:00:05.000 --> 00:00:05.000
>>
>> (MEN CHEERING)
>>
>>
>> 00:00:05.500 --> 00:00:10.500
>>
>> You guys wanna story?
>>
>>
>> 00:00:05.500 --> 00:00:10.500
>>
>> (MEN CHEERING)
>>
>>
>> 00:00:11.500 --> 00:00:15.500
>>
>> You guys wanna story?
>>
>>
>> 00:00:11.500 --> 00:00:15.500
>>
>> (MEN CHEERING)
>>
>>
>>
>> Regards,
>>
>> Boy

Received on Friday, 12 November 2021 16:21:31 UTC