Re: Captioning for live video

On Mon, Dec 17, 2012 at 8:01 PM, Cyril Concolato <
cyril.concolato@telecom-paristech.fr> wrote:

>  Hi Silvia,
>
> Le 12/15/2012 12:14 PM, Silvia Pfeiffer a écrit :
>
> Hi Cyril,
>
> On Fri, Dec 14, 2012 at 10:03 PM, Cyril Concolato <
> cyril.concolato@telecom-paristech.fr> wrote:
>
>>  Hi Silvia,
>>
>>    2. Solution in WebVTT
>>>>
>>>
>>> Separately from WebVTT syntax issues, currently the spec does not
>>> support streaming of WebVTT in <track>.
>>>
>>
>> That's what https://www.w3.org/Bugs/Public/show_bug.cgi?id=18029 is
>> about. I continue to believe that similarly to how we can stream a video
>> live to a <video> element, we should be able to stream text to a <track>
>> element and synchronize it from the cue time stamps.
>>
>>  I fully agree with that. In this respect, you might be interested by the
>> demo I made at MPEG back in October, showing synchronized playback of
>> streamed subtitles and video in a live scenario using DASH. Details are
>> here:
>>
>> http://concolato.wp.mines-telecom.fr/2012/12/14/live-streaming-of-video-and-subtitles-using-dash/
>>
>
> Ah this is very interesting.
>
> I wasn't able to find this by reading your slides and the link to the demo
> goes to a file on C:// : did you use JavaScript and TextTrack to feed
> WebVTT cues to the browser for rendering?
>
> The demo indeed used C:// as this was using the meeting's network which
> was not so good and also because it's hard to let a live server run
> permanently on our university servers. I'm working on that to try and make
> the demo available online.
>

Fair enough. I'm keen to try it out, so do share the link when you've got
it available publicly. Thanks!


 Yes, to feed the WebVTT cues, I followed a suggestion from Philip
> Jägenstedt [1]. I have a dummy HTML video element used to load the WebVTT
> segment and when it's loaded, I transfer the cues in JavaScript to the real
> video element.
>

Did you need that because you were using DASH with full WebVTT files for
every segment? Or simply because you are delivering multiple cues in a
single XHR?



> I would have preferred either a JavaScript function to parse the WebVTT
> segments
>

Since cues are separated by empty lines, wouldn't it be simple to write a
cue segmenter and then feed the individual cues to the addCue() function?



> or even better, support for WebVTT in Google's MSE implementation.
>

You're referring to this spec, I assume:
http://dvcs.w3.org/hg/html-media/raw-file/tip/media-source/media-source.html

How would you add it there? I think we may need to work with the MSE group
on this if that is indeed the way to go.



>  Apparently, one way to do that would be embed WebVTT in MP4 or WebM, but
> that's a bit awkward I think.
>

Embedded is already supported in the spec (even if implementations are
still catching up).

Silvia.