|
Honors:
NSF research grant (2004-2007); Arnold O. Beckman Research
Award (2004); Henry Rutgers Fellowship (1989); NSF research
grant (1991-1992), Critical Research Initiatives (2005-2008).
Research Interest:
Prosody conveys meaning beyond words. A human speaker controls
prosody by changing duration, f0, intensity, among other acoustic
attributes. Shih's work focus on prosody representation and
modeling. The goal is to generate prosody using representations
and parameters that match human perception and production
of prosody, and to generate a full range of natural prosodic
variations that are suitable for speech technologies such
as text-to-speech systems and automatic speech recognition
systems, and for language teaching.
Shih has completed an articulatory-based prosody modeling
platform Soft Template Markup Language, which functions as
both a prosody generation system and a data-driven machine
learning system. The model has been applied successfully to
capture stylistic variations in natural speech and in singing.
Shih's current research projects include the prosody of disfluency,
focus realization, cross-linguistic prosodic perception, representation
and modeling, and the implication to fluency in second language
acquisition.
|
