Sixth International Conference on Spoken Language Processing (ICSLP 2000)

Beijing, China
October 16-20, 2000

On the Use of Speaking Rate as a Generalized Feature to Improve Decision Trees

Robert Faltlhauser, Thilo Pfau, Günther Ruske

Institute for Human-Machine-Communication Technische Universität München (TUM), Germany

Decision trees are probably the most common way for gen- erating models for phonemes in their phonetic context. In this paper we investigated several ways how speaking rate information can be integrated in the decision tree process. We basically focused on two approaches: on the one hand a speaking rate feature included in the decision tree itself and on the other hand a pruning approach for creating individual model sets. Recently, some papers have come up with the idea to include a gender feature already in the decision process. In our paper we went a step further and wanted to see whether speaking rate can be a fruitful extension to decision trees. Experiments have shown that the introduction of speaking rate leads to improvements in combination with a general gender feature. Further experi- ments with di erent pruning strategies aimed at creating adequate model sets for di erent speaking rate categories.

Full Paper

Bibliographic reference.  Faltlhauser, Robert / Pfau, Thilo / Ruske, Günther (2000): "On the use of speaking rate as a generalized feature to improve decision trees", In ICSLP-2000, vol.1, 317-320.