TTS-driven Synthetic Behavior Generation Model for Embodied Conversational Agents - Coverbal Synchrony in Human-Machine Interaction - page 353

adverb-pronoun (CU-1) and particle-pronoun-noun (CU-2). For each
meaningful phrase, the system creates a corresponding Content unit
within the ContentUnits relation layer (CU-1 and CU-2 in Figure 11).
In this case, the meaningful word-type in both meaningful phrases is
identified as a pronoun. The meaningful word also provides the same
meaning in both cases. As shown in Figure 11 (conversational shapes),
the meaningful word is described by the meaningful hand shape
right_K7, and arm configuration named Fr|Ab|Cl|O (front, abdomen,
close, orthogonal). The most probable initial state of the meaningful
shape is right_B1 for handshape, and Fr|Ce|No|O (front, chest,
normal, orthogonal) for the arm configuration. Then follows the inner
fluidity deque . At this level the data stored in the Phrase and Syllable
relation layers can be used in order to adjust (if necessary) the extent
of each CU, and in order to create a movement plan (structure) for the
propagation of shapes within each CU. The Phrase layer in this case
contains the following predicted phrase breaks: Vedno(*) iste(B2), (*)
in(*) samo(*) iste(*) obraze(B3). As a result, the CU-2 item is extended
(in order to also include the word 'in'). On the Syllable relation layer,
the inner fluidity deque searches for emphasized syllables—those
including syllables with PA (primary accent) and NA (secondary
accent) labels. The syllables labeled as PA represent utterances over
which the stroke is performed. In the system, the stroke phase will
be co-aligned with the beginning of the PA-word, and the ending of
its PA-syllable. In Figure 11, the CU-1's stroke phase starts with the
beginning of the word 'iste' ( the same ), and ends on the phoneme 's'.
During this interval, the ECA will perform transition from shape,
identified within the Phase item PI-1 (right_B1, Fr|Ce|No|O), to
the shape identified by PI-2 (right_K7, Fr|Ab|Cl|O). Similarly, the
shape transition from PI-3 to PI-4 (first image of the PI-4 sequence) is
Figure 11. Transforming HRG structure into EVA-SCRIPT-based procedural animation
specifi cation.
(Color image of this fi gure appears in the color plate section at the end of the topic.)