Russell Shepherdson www:http://www.computing.edu.au/~shepherd email: shepherd@cs.curtin.edu.au

As far as facial gestures are concerned, the FAITH talking head (TH) does nothing but move its mouth when it speaks to the user. The TH needs to exhibit behaviour that is temporal, dynamic and has the lifelike, somewhat unpredictable feel of human behaviour. The outcome of the research work is an emotionally expressive talking head with a personality and mood consistent with the author's goals and intentions. Implementation: From our research, we have concluded that the facial expressions and gestures in the following table contribute to the human perception of personality. To allow for the specification of a specific personality, we have provided a set of high-level parameters (for each facial feature) that sit on the facial animation parameter (FAP) set. For example, the parameter set for head turning includes:

• the max and min speed for each head turn
• the variance of the head turn speed
• the max and min values of head yaw (these directly relate to the head yaw FAP value)
• the max and min values of head pitch (these directly relate to the head pitch FAP value)
• and the number of frames to do a complete head turn.

Yehuda Levy www:http://www.computing.edu.au/~levyy email: levyy@cs.curtin.edu.au

The current implementation of FAITH runs either locally, or through a dedicated client-server application. This requires the installation of the client on the end-users' machine, severely limiting the widespread use of the application. The current global communication environment is the Internet, the Web in particular. Unless FAITH will be able to run inside an ordinary Web-browser, it will never reach the mass audience it deserves, nor be able to integrate with the content of the Web, i.e. as a customer-service agent or a virtual salesperson.

FAITH is currently platform-dependant. To run on the Web it needs at least a platform-independent interface to the browser, such as Java. However, in order for it to run fastest and with minimal code re-writing, the use of native code is important. A trade-off exists between the use of Java and native code.

The TH runs in a GL drawing area, implemented inside a Java applet. The GL rendering and 95% of the application is still implemented in native code. To achieve this, GL4Java is used as the interface for obtaining the native GL area from from an applet. Only the minimal GL4Java code is used (based on GLContext), which is converted to a DLL (for Windows). 5 other DLLs are created from a stripped-down version of FAITH. The 6 DLLs, zipped together with the HTML and the Java applet, can be automatically downloaded and installed from the HTML code. Once installed they can be used for playing any TH file.

John Stallo www:http://www.computing.edu.au/~stalloj email: stalloj@cs.curtin.edu.au

John Stallo is enrolled as an Honours student in the Department of Computing, at Curtin University of Technology. He is interested in simulating the effects of emotion on speech for the Text-to-Speech module of a Talking Head. A significant portion of the project involves implementing an XML-based speech markup language. The markup language will have low-level tags to control speech parameters such as pitch, rate, emphasis etc. The simulation of a particular emotion will be specified via higher level tags.

The main aims of the project are:

1. To develop a system that can add simulated emotion effects to synthetic speech. This involves researching the speech correlates of emotion that have been identified in the literature. The findings will be applied to the control parameters available in a speech synthesizer, allowing a specified emotion to be simulated using rules controlling the parameters.
2. To integrate the system within the text-to-speech module of the Talking Head being developed as part of the Interface project.

Project Quick-Links:

• Main Honours Project Page
• Honours Project Proposal
• Speech Samples
• via Interface site
• via my Homepage

Bernhard Tschirren www:http://www.computing.edu.au/~tschirre email: tschirre@cs.curtin.edu.au

Previous Techniques all plant roots and grow individual hair strands.
Two significant problems:

• Too Much Data: Head has 100,000 to 150,000 hairs. If each hair were simulated by a chain of 32 cylinders, and each cylinder were represented using 32 triangles, then hairs would have to be modeled using 102,400,000 triangles. To render this amount of geometry at 15 frames per second, a system would need to achieve at least 1536 triangles per second.
• Aliasing: Each hair strand is typically less than a pixel wide. The solution is to use some kind of super-sampling technique. However, this makes the hair look unusually smooth and flat.

1. Geometry Generation: The geometry generation involves producing vertices, faces and texture coordinates, such that they represent a hairstyle. The hair surface need not be flat and continuous. A better solution would be to create a set of interwoven faces, that when textured appear to be solid. For example, trees in a landscape simulator are usually modelled using two intersecting rectangular faces. Each face is textured by a profile view of the tree. The result is the appearance of a three dimensional tree, without actually modelling all the details of the tree.
2. Texture Generation: Produce a set of texture maps that can be applied to the geometry. Created using procedural techniques that can reproduce the painters effect. They must capture the essence of hair, while being very low resolution. Research into art and painting techniques is looking very promising.

Quoc Hung Huynh www:http://www.computing.edu.au/~huynhqh email: huynhqh@cs.curtin.edu.au

The above sited project is to be supervised by Andrew Marriott, and will involve the investigation of a virtual news presenter (VNP) package to allow an animated floating head to present information supplied from multiple sources, such as text files in a typical news presenter manner. The VNP package will allow the output of video data such that it can be either pre-recorded or displayed live onto existing broadcasting hardware. The VNP package is to be developed in conjunction with the school of vision and learning and channel 31 broadcasting. It is foreseen that the VNP package will provide the mechanism for channel 31 broadcasting to present factual information on their broadcasting network in a unique and innovative manner.

The project will involve the investigation of current facial animation technology and text to speech synthesis (TTSS), including the current work developed by the Computer Animation Negus (CAN), specifically the talking faqbot application. This project will work in conjunction with other similar projects dealing with facial animation. Typically, the supplementary projects will look at increasing the quality of TTSS, quicker rendering time and the addition of a synthetic character persona.