Draft Timed Text requirements

Pierre and I had a chat about organizing the various ideas from last week into a coherent message. Here's my draft write-up after our conversation. (Pierre hasn't reviewed this, yet.)



Timed Text translation problem statement:
With the two independent W3C Timed Text specs today (TTML & WebVTT), Timed Text is often be authored in one spec but rendered on a client supporting the other spec. This requires a translation between the two specs. With no defined mapping from one to the other, we risk mistranslating or dropping authored Timed Text information when it is rendered on the client. This results in a poor user experience and could also result in violations of government regulations on Timed Text display. One likely workaround will be that authors will use a least common denominator approach to keep within a safe, translatable subset between the two specs. But this will mean that end users miss out on the richer capabilities of both specs. For example, in the US, captions could continue to be authored at a simple EIA-608 equivalent to avoid translation issues, which cheats end users of the richer experience possible today.

The Web & TV IG recommends the following requirements for W3C Timed Text efforts:

1. The needs of timed text authors and timed text consumers should be prioritized over the needs of specification writers and implementors.

While the focus of W3C efforts is often on authoring Web specifications and implementing Web client code, Timed Text needs to be viewed as an end-to-end ecosystem where the Timed Text is created as part of media authoring, often outside the Web, and the Web is one of many clients. The W3C Timed Text specifications and strategies need to integrate into this larger context.

2. The overriding goal should be to maximize the consistency between the authored Timed Text and the rendered Timed Text.

Media content augmented by Timed Text is very vulnerable to even small mistranslations. For example, it's easy to think of movies where the loss of a single line of dialog would distort the understanding of the entire movie.

3. We recommend that the Timed Text groups work together towards one or more strategies that would maximize the consistency between Timed Text authoring and Timed Text display. We suggest three possible logical strategies towards this goal, but make no recommendations about which of these strategies is most practical or preferred.

3a: No translation via a single spec. The W3C could work with media standards groups on a strategy of driving a single Timed Text spec for the end-to-end media ecosystem. This could be based on TTML, WebVTT or some new merged effort. All authors and clients implementing the same spec would clearly eliminate translation errors.

3b: No translation via clients supporting both specs. The W3C could evangelize that all media clients support both client specs, much like Web clients support both PNG and JPEG. Again, this would eliminate all translation errors.

3c: Define full mappings between the client specs. By defining the two client specifications and how to map between the two, translation errors can be minimized or even eliminated.

4. Any of the above strategies will require minimizing the profiles clients must support. For example, defining a common, limited-resource client definition that will work across national variants will be critical for both specs.

Received on Thursday, 30 May 2013 15:01:21 UTC