Glossary of Terms

Following is a list of terms specific to either the DIVA project or the present discussion, and their definitions.

Accent - a set of files based upon training sessions which contains information regarding each user's specific gestures. Accent files specify "weights" which are used in the calculation of phoneme probabilities from input glove & tracker data. "Accents" is also the interface in the DIVA project used to create accents.

Artisynth - a 3-d modeling toolkit written in Java. The software runs a main window, and allows the user to choose from a set of models to load. This is the software from which the Kura Face model, the talking face used in DIVA, is run.

Dictionary File - in the DIVA project, a file which contains a list of different expressions. Multiple expressions are not supported in the present state of the project, and represent a goal of future work. Each expression represents a particular facial emotion, and applies to entries in the map file, such that for each phoneme, each expression must map to a particular PC vector.

Expression - a particular facial emotion - each dictionary file specifies a list of expressions.

Eclipse - a powerful Java development environment for managing and building projects, running and debugging, generating documentation, and other tasks. Can be downloaded from the web -- a working copy is located on Brahms at BrahmsOSX:/Eric/eclipse with workspace at BrahmsOSX:/Users/diva/Documents/workspace.

Java - software development language used to code the Artisynth project, and MXJ objects.

KuraFace Model - a 3-d deformable face mesh which exists as a class in the Artisynth project. The face consists of a set of nodes which can be stretched using eight parameters, called Principal Components.

Map File - in the DIVA project, the file used to specify a mapping of phonemes to facial expressions, which is necessary to drive the talking face. Each combination of phoneme & expression (ex: EE happy, EE sad etc.) is associated with a particular vizeme file, which specifies a PC vector and thus a particular facial shape for that entry.

MAX - programming language used to code the DIVA project. MAX is a visual language, primarily used for audio processing, which is based upon objects connected with inlets and outlets connected by "patch cords". Data flows through patch cords from outlets to inlets and is processed by the objects.

MXJ - Java-coded objects for MAX/MSP. The package of classes com.cycling74.max contains an implementation for MAX objects with methods for inlet and outlet behaviour which can be overwritten.

PC Vector - the Principal Component Vector, a list of 8 values which specifies the stretching parameters for the KuraFace. A PC Vector of 0 0 0 0 0 0 0 0 sent to the face yields the neutral or unstretched face -- changing any one of these values will cause the face to stretch in one of eight different ways. A particular facial expression of the Kura Face is thus specified by a unique PC vector.

Phoneme - any one of the auditory building blocks of language, or a single particular sound that arises in speech. Examples are the "eeee", "aahhh", "ffff" sounds.

Plosive - one of the eight abrupt or "stopped consonant" sounds, p k t ch b g d j.

Radial Basis Function - for detailed explanation, please look up in the "Publications" page in the menu to the left. RBF's use accent weights to calculate phoneme (or in Future Work, expression) probabilities from performance glove & tracker data.

Session File - file type saved by the training window, which contains data for recorded gestures.

Training - interface in the DIVA project used to record gestures in session files. This is how the user specifies to the computer which specific gestures they will use for particular phonemes or expressions.

Vizeme - the visual analog of a phoneme -- the "EE" vizeme is the appearance of a face when it is making the "EE" sound, and similarly for all other phonemes.

Vizeme File - in the DIVA project, a file which contains a single PC vector, which specifies a particular facial expression. Vizeme files are managed by the VizemeBuilder interface, and each vizeme file is associated to a particular phoneme, such that there is a set of "EE" vizeme files, "I" vizeme files, "F" vizeme files, etc. The vizeme files represent a bank of possible facial expressions which can be used when constructing a phoneme --> PC vector mapping in the CreateVizemeMap interface.

Vowel/Consonant Decision Variable - a value between 0 and 1 which specifies the relative strength of vowels or consonants. In performance mode, the true vowel or consonant probability is determined by multiplying the RBf-determined phoneme probability by the vowel/consonant decision variable.