Section: New Results
Document Formats and Description
The XTiger language (see section 3.4 ) created in the team in 2006 is versatile enough to represent templates that can capture the overall structure of large documents as well as the fine details of a microformat  . It borrows from various types of languages found in the XML family, such as schema languages or transformation languages. It is different from these languages however, because its main goal is to drive the creation of new documents instead of validating or transforming existing documents. This brings original features to the language, in particular its ability to be emmbedded in the documents it describes, instead of constituting external resources.
The audio part of JSR-234, for which reference implementations exist, enhances the audio support on mobile devices by adding rendering features like 3D audio and virtual acoustics, but also by introducing a 3-level hierarchical model for mixing, muting and rendering parameters, thus allowing interesting special audio effects. At this point, it is interesting to observe that iXMF is a format with no API and that, at the opposite, JSR-234 is an API with no format. We have defined an XML format (iXMFT) for JSR-234 which could be considered as a Tiny iXMF format. An editor and its associated soundtrack manager are in construction using EMF databinding technology (see section 6.3 ).
The activities with INA through the PhD thesis of Marc Caillet on multimedia content description have provided two main results in 2006:
The FDL language (FERIA Description Language) has been completely recast. The goal was 1) to obtain a meta-model that handles homogeneously instances and classes hierarchies and 2) to make it possible to specify a temporal order among temporal descriptors. Moreover, a whole document model has been specified and implemented to handle not only media assets but also document collections. Different tools for managing FDL descriptors have also been implemented: a parser for the production of internal structures, a descriptor database for storing them, and a temporal request interpretor for accessing them.
An application called JAM! (Jouons Avec le Misanthrope) is being experimented with FDL. Thanks to a cooperation with François Yvon from ENST, a corpus of six performances of this play is available for segmentation: several kinds of structural segmentations are performed on audio (phonems, words), text (acts, scenes, verses) and video (partial scene segmentation) with synchronisation between them. These sets of descriptors pave the way for the development of several navigation and publication applications. The objective is to demonstrate how FDL can provide generic behaviors that can be attached to descriptors classes and reused by other descriptors, thanks to their relations in the class hierarchy.