Automatic Corpus-based Tone using K-TOBI Representation

In this paper, we present a prosody generation axchitecture based on K-ToBI (Korean Tone and Break Index) representation. ToBI is, a multi-tier representation system based on linguistic knowledge to transcribe events in an utterance. The TTS system which adopts ToBI as an intermediate representation is known to exhibit higher flexibility, modularity and domain/task portability compared with the direct prosody generation TTS systems. However, the cost of corpus preparation is very expensive for practical-level performance because the TbBI labeled corpus has been manually constructed by many prosody experts and normally requires large amount of data for statistical prosody modeling. Contrary to previous ToBI-based systems, this paper proposes a new method which transcribes the K-ToBI labels completely automatically in Korean speech. We developed automatic corpus-based K-TOBI labeling tools and prediction methods based on several lexico-syntactic linguistic features for decision-tree induction. We demonstrated the performance of F0 generation from automatically predicted K-ToBI labels, and confirmed that the performance is reasonably comparable with state-of-the-art direct prosody generation methods and previous TOBI-based methods.