Abstract
This work presents the design of a distributed sign language synthesis architecture. The main objective of this design is to adapt the synthesis process to the diversity of user devices. The synthesis process has been divided into several independent modules that can be run either in a dedicated server or in the client device. Depending on the modules assigned to the server or to the client, four different scenarios have been defined. These scenarios may vary from a heavy client design which executes the whole synthesis process, to a light client design similar to a video player. These four scenarios will provide equivalent signed message quality independently of the device’s hardware and software resources.
TopBackground
Literature provides several examples of SL synthesizers. In order to represent synthetic signed messages, two main techniques have been developed:
- 1.
The first approach to SL synthesis consists of creating a composition of small segments of video (Solina, Krapež, Jaklič, & Komac, 2001). This approach to SL synthesis requires image processing and a great number of pre-recorded sequences in order to act as a synthesizer, and thus significant storage capacity.
- 2.
The second main approach to SL synthesis uses virtual avatars. H-Anim (ISO/IEC 19774, 2005) is the most widely used avatar structure; it is a standard definition for human representation on VRML (ISO/IEC 14772-1, 1997) or X3D (ISO/IEC 19775, 2004). Within avatar animation category, there are two different approaches related to the definition of the animation. The first one uses continuous motion data obtained from (a) an expert signer using different motion capture techniques or (b) manual animations created by an expert animator. Although the results obtained with this technique are natural, Kennaway (2002) described several disadvantages of this approach based on the difficult adaptation of the recorded data to avatars with different anatomies. The second approach to define the animation for avatar-based SL synthesis uses a parametric definition of the signs in order to generate the animation (Bangham, Cox, Elliot, Glauert, & Marshall, 2000; Irving & Foulds, 2005; Kennaway, Glauert, & Zwitserlood, 2007; Zwiterslood, Verlinden, Ros, & van der Shoot, 2004). The resulting avatar animation is not as natural as the one obtained using the continuous motion data approach. However, the animation quality is the same over the whole sentence and the storage requirements are highly reduced. The parametric synthesis is the only approach that provides enough flexibility to define all the SL linguistic variations.
Key Terms in this Chapter
Remote Rendering: The 3D scene processing is carried out on a remote server and the results are sent as a video or an image to the client device which performs the visualization of the results.
Signing Avatar: Virtual anthropomorphic character used to represent signed messages. Its animation can be manually defined, captured from a human signer, or parametrically described.
Ubiquitous Sign Language Synthesis: Generation and access to artificial signed messages regardless the hardware and software resources of the client’s device and the quality of the network connection.
HLSML: XML-based notation focused on signed message definition and its prosody; it does not require describing the phonology of the signs, as other notations do.
Remote Gesture Synthesis: The avatar animation definition process is carried out on a remote server, the obtained animation tracks are sent to the client using a pre-established format.
Parametric Sign Language Synthesis: Generation of the synthetic signed message based on phonetic-like sign definitions, reducing the storage requirements and incrementing the linguistic flexibility.
Modular Architecture: Distribution of a complex application into several and more simple units that can be run in different devices. The communication between these units is done using a predefined interface.