Fourth ISCA ITRW on Speech Synthesis

August 29 - September 1, 2001
Perthshire, Scotland

Predicting phrase breaks with memory-based learning

Bertjan Busser, Walter Daelemans, and Antal van den Bosch

ILK / Computational Linguistics, Tilburg University, The Netherlands

We investiate whether Memory-Based Learning (MBL) can be used to predict Phrase Breaks (PBs) in speech production reliaibly. The MBL approach is compared to tha HMM approach described in [Taylor and Black, 1998] using the same corpus and information sources. We show that a simple memory-based learning algorithm that uses only minimal context and inforrmation outperforms tha HMM approach, in turns of precision and recall, An exhaustive search of variants of algorithms, metrics and information sources does not bring any significant further improvement.

Full Paper

Bibliographic reference.  Busser, Bertjan / Daelemans, Walter / Bosch, Antal van den (2001): "Predicting phrase breaks with memory-based learning", In SSW4-2001, paper 125.