[Matroska-devel] Re: Re: EBML (subitles)

[unmei]

Martin Nilsson wrote:

> The reason why I mentioned subtitles in the first place is beacuse 
> there is some sort of interpretation of the formats when separating 
> them into Track->CodecPrivate and Cluster->Block. Since there is 
> interpretation, why not also have normalization, or at least 
> "compression" to a binary format? One could note that the segmentation 
> of SSA files didn't help us with our video player, since our SSA 
> decoder worked on complete SSA files. Skip to specific words in the 
> subtitles are also a more costly operation now that you have to seek 
> through more data.
>
> But I guess that I am talking for no reason. The "correct" solution 
> from Matroska point of view is for me to present a new format and hope 
> for its adoption.
>
sorry to jump into this discussion a bit, but i snapped up the subtitles 
parts :)

i'm currently developing a new graphical subtitle format. It's in the 
early implementation and still change phase but it shows good results so 
far. I hope it will allow to compress full color subtitles at about the 
size of (4 color) sub/idx. I think 1 mb per stream (language) is 
possible and maybe even 400 or less kb could be attained. I designed it 
to be not CPU demanding by not incoprorating sophisticated compression 
schemes (_not_ like png, lzw et all) but my current encoder is extremely 
slow (probably because i re-resize the target array quite often and 
hopefully more avoidable stupidities).

What i actually wanted to stress is the format is very heavy on the 
header, about 90% of all data is in the header and must be decompressed 
before playback. So for me its nice to hear players prefer to handle 
subtitles as one big chunk and do not rely on too much on "streaming" them.