Preliminary work on speech unit selection using syntax-phonology interface

This paper proposes an approach which uses a syntax-phonology interface to select the most appropriate speech units for a target sentence. The selection of the speech units is done by constructing the syntax-phonology tree structure of the target sentence. The construction of the syntax-phonology tree is adapted from the example-based parsing of UTMK machine translation. In the process of constructing the syntax-phonology tree, we first identify the related trees from the speech corpus. Then, the generated subtrees based on the related trees are combined. The concatenation of the combined subtrees nodes is the synthetic utterance of the target sentence.