I. Bulyko, Flexible speech synthesis using weighted finite-state transducers, Ph.D. thesis, Electrical Engineering, University of Washington, Seattle, WA, Mar. 2002.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....representing additional sources of information can be chained in series. For example, duration or lexical stress could be another diacritic to be matched by the unit selection search along with pitch. An independence assumption underlies this additive chaining of costs. In work by Bulyko [21, 20], simplified ToBI [131] prosodic markers as part of unit selection. 138 1 2 # # L3 R2 Figure 5 20: FST of corpus utterance diacritics. can be added around units to describe intonational or paralinguistic attributes. In this example, the belongs to classes of ....

I. Bulyko, Flexible speech synthesis using weighted finite-state transducers, Ph.D. thesis, Electrical Engineering, University of Washington, Seattle, WA, Mar. 2002.

....units, the resulting clusters may overlap in the acoustic space covered even though they do not overlap in terms of set membership. In synthesis, such overlapping clusters may be desirable for finding the best unit sequence, to the point of introducing sharing of some units in multiple clusters [12]. 3.3. Speech Models Hidden Markov models have been used more directly in speech synthesis in two ways: as a model on which to assess or reduce target and concatenation costs, and as a generative model for the actual synthesis process. In concatenative speech synthesis, the output quality ....

I. Bulyko, Flexible speech synthesis using weighted finitestate transducers, University of Washington, Ph.D. Dissertation, Electrical Engineering, 2002.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC