Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Learning and Transfer of Learning for Synthetic Speech

Martine van Zundert, Jacques Terken

IPO, Center for User-System Interaction, Eindhoven, The Netherlands

Understanding synthetic speech involves a learning process. We address the question whether transfer of learning takes place from one kind of synthetic speech to another. An experiment is presented in which learning curves for intelligibility were determined for two diphone-based synthesis systems for Dutch, A and B, with different diphone databases. Twenty-four subjects heard eight blocks of 50 Semantically Unpredictable Sentences (SUS). Four different experimental conditions were constructed: In conditions AA and BB, the same synthesis system (A and B, respectively) was presented in blocks 1 to 4 and 5 to 8. In conditions AB and BA one system was presented in blocks 1 to 4 and the other system in blocks 5 to 8. Results show that learning effects are observed within systems (conditions AA and BB). However, we find no evidence of transfer of learning between systems (conditions AB and BA).


Full Paper

Bibliographic reference.  Zundert, Martine van / Terken, Jacques (2000): "Learning and transfer of learning for synthetic speech", In ICSLP-2000, vol.3, 949-952.