Live News Broadcast

Broadcasters can currently classify programmes in their schedule using more traditional metadata. By maintaining records including the time of a

programme, its name, presenters and a brief overview, viewers can access this data through one-way mediums encoded within the analogue television signal, such as teletext. More recent Digital Video Broadcasting (DVB [62]) can present this information within an Electronic Programme Guide (EPG), presenting the user with “now and next” information. As and when schedules are changed, this data can be pushed to the user over normal broadcast channels.

With the advent of digital personal video recorders (PVRs), random access storage of media is available into the home, and broadcasters are beginning to implement simple content classification systems in addition to schedule reporting [61]. This allows the recorder to track the types of programme a user watches, and infer that it should speculatively record other programs of a similar type, through use of an agent based recommender system [47, 50]. The classification of programmes is often performed by hand, limiting the amount of markup applied, although coarse-grained automation has been developed [143].

In these current systems, classification and markup must be performed before the programme is broadcast, and applied at a granularity which is the length of a programme; no knowledge is transferred about sections of material within the programme.

Through use of continuous metadata, information of relevance within, and throughout, the programme can be sent to the user. This is particularly true in the case of live programmes, when although there might be extra information available, the exact detail of content within the programme is only known a short time in advance. Rolling news channels are a good example of this.

Using continuous metadata flows that are separate from the video stream, the main video could be sent to the user with traditional broadcasting techniques. The metadata flow could be delivered using a lower bandwidth internet

connection (cable modem, x DSL etc.) to the home, and then to a suitable display device within the home. This might be integrated into a television, or more likely a PC, or wireless handheld device. Where aerial or satellite reception were not available, for example programmes from international or minority broadcasters, the video stream might also be sent using internet protocols; one user might watch programmes mixed from several broadcasters received in a multi-modal manner.

Any programme content classification could now be generated dynamically as the programme was filmed live, and transmitted on a separate continuous metadata flow. For example, a user might choose to receive the “programme description” continuous metadata flows from multiple broadcasters while only watching a single media data stream, which is constrained by bandwidth (for Internet streams) or equipment limitations (for traditional television receivers). The display device could monitor the content metadata for all programmes currently being broadcast, and switch the video data to prioritise a channel broadcasting, for example, sports bulletins.

Links to relevant web or hypermedia documents also constitute metadata, and can be delivered as a continuous metadata flows [110]. The broadcaster of live television news might provide a flow of links to documents on their website, corresponding to the current news item, or to interactive discussion boards and messages within them. Following a link might pause the temporal display of video stream, recording it to a PVR, for time-shifted playback after reading the document.

Links are not limited to referencing non-temporal documents, but could also begin streamed playback of archived video footage, programmes or reports. These in turn would have their own archived continuous metadata flows, in addition to those running live.

Distribution of continuous metadata sources would enable the broadcaster to franchise their media stream, allowing other news commentators to generate link flows containing their own links and comment. A user could select different

suppliers of augmentation according to their own taste and preference, rather than that of the television company. Communities of like-minded viewers might even form to share insight, using webs of personal metadata flows.

A user might also run a recommender system locally, to analyse link trails previously followed, and select the best suited links from the many incoming metadata flows (or perhaps create a new link flow source based on that users’ browsing history).

Subtitles, which are normally raster images superimposed over a video stream, are essentially a text augmentation, which could also be separated into

continuous metadata. The television broadcaster might provide a subtitle flow in the native language of the programme, typed live along with the broadcast (and with advances in technology, perhaps automatically generated or assisted).

Using distributed continuous metadata sources, translators could provide flows of subtitles in other languages, independently from the original broadcaster. One agency might provide translations for several television channels, using the same infrastructure. This could be performed manually, with translators using the original video stream; or, since the primary subtitle flow is already text based, basic translation could be automated.