Timed Text and Subtitling

Potential users of timed-text are still undecided on which format to use. John Birch, Strategic Partnerships Manager, Screen Systems, discusses the various market areas and which standards to employ.
The term 'Timed Text' is typically used to identify documents containing subtitles or captions for use with Internet based video. These documents unsurprisingly contain text and timing data and optionally styling information and metadata. Many different formats of timed text document exist, supported either by user communities, commercial organisations or by formal standards bodies (e.g. SRT, WebVTT, SAMI, SMIL and TTML/DFXP).
Unfortunately, the interpretation of timed-text documents, and hence the appearance of the text content, can differ between players and between devices, with positioning and font style particularly inconsistent. The majority of timed text formats have been developed to satisfy accessibility demands for closed-captions, and this requirement is generally met. However, for translation subtitling, the limitations of many existing timed text formats present a problem and the resulting viewer experience may be well below that of broadcast or cinematographic subtitles.
To address the generic issue of specifying how to display text content synchronised with other media elements, the W3C established the ‘Timed Text Mark-up Language’ working group, which started work on the 6th March 2003. The first draft of a proposed formal standard (called TTML-DFXP) was released in October 2005 and this finally became a recommendation in November 2010. A revision to this recommendation is expected soon. A parallel effort by the SMPTE digital cinema group also released a formal standard targeted at cinema subtitling in the same time frame (SMPTE 428-7). Both the W3C and SMPTE standards are XML based, allowing for extensibility and making subsequent revisions easy to manage.
The TTML standard includes a profiling mechanism that allows software to determine if it can display or process a document correctly. One profile, the TTML Simple Delivery Profile for Closed Captions (US), has been developed by W3C specifically to facilitate delivery of captions for online video presentations. The profile minimises development time for implementers as they only need to create support constrained presentations.
Other standards organisations have built upon the TTML framework. SMPTE have defined the ST 2052-1:2010 profile specifically for use in an end-to-end broadband infrastructure as a native format.
This profile adds broadcast-specific metadata and extends TTML by adding support for PNG format images and a binary data carriage mechanism. This allows CEA 608 captions to be ‘tunnelled’ through SMPTE-TT and recreated at the end-device. Importantly, the SMPTE-TT profile sets explicit initial conditions for white text displayed centred at the bottom of the screen. The SMPTE group have also produced recommendations for converting existing caption and subtitle data (e.g. 608, 708 and 428-7) into SMPTE-TT (ST 2052-10, ST 2052-11, ST 2052-12 respectively).
The Digital Entertainment Content Ecosystem (DECE) has adopted the SMPTE-TT profile for the Common File Format (CFF) Specification used in the Ultraviolet cloud-based licensing system. The subtitle tracks are treated like audio and video media types and subtitle “samples” are synchronised with other media tracks selected for presentation. The CFF-TT specification defines two profiles – one for text and one for image based subtitles, but mixing of text and image subtitles within one CFF-TT document is not supported. Significantly, CFF-TT defines a number of general restrictions on the use of TTML-DFXP vocabulary and grammar within CFF-TT documents.
In a similar manner, the EBU established a working group to specify an XML based format for archiving and interchange of subtitles to replace the currently widely used EBU ‘STL’ format (Tech 3264). The EBU group intend to publish a series of documents to supplement the published format definition (Tech 3350), called EBU-TT, which is also a subset of TTML 1.0. Planned documents include a guide to converting EBU STL documents to EBU-TT, a new protocol for live subtitling that is compatible with EBU-TT, an enhanced version of the new format that can be used in future scenarios for ‘authoring of intent’ and a general user guide.
The enhanced version of EBU-TT is intended to allow a more abstract form of a subtitle archive file which could be machine-converted to different formats of subtitle file, each targeted at a specific use case.
The addition of extra metadata allows the inclusion of context and information about the associated video content (e.g. description). As an example, including information about speakers would allow a conversion to automatically assign colours, italics or speaker change marks when generating Teletext or CEA 608 captions respectively. Like CFF-TT, the EBU Timed Text format recommends that attributes and their values are explicitly specified in documents, rather than relying on their default TTML values.
The majority of timed text formats supported by formal standards bodies are XML based and reference the original TTML specification, but there remains a question as to which format an organisation should choose. For distribution, the choice is often dictated by the customer, but for archive purposes it is suggested that an extensible format that allows the subtitles and captions to be stored as ‘rich content’ is better suited to future re-use and re-purposing. In general, it is possible to convert between all the timed text formats, but clearly, if the necessary information is not, or cannot be stored in the ‘master’ archive format, it is far harder to add it later.
We are aware that various bodies are working on making these standards interchangeable and we offer encouragement to those involved in trying to achieve common ground.
As it currently stands, however, Screen is able to offer full support for whichever option you choose.
Read the article in the online edition of Regional Film and Video here.