7th International Conference on Spoken Language Processing
September 16-20, 2002
In this paper we conduct an evaluation as well as a reassessment of the PhonDatII spoken language resource. 10 years after the record of PhonDatII it is time to summarize and to look into its future. At present, the corpus comprises 39612 manually labelled phone tokens and 15083 syllable tokens of read German utterances. We describe the corpus in detail, and then we present a new method to evaluate segmentation boundaries. Finally, we ask the question as to how we can refine the PhonDatII database for the future. The mean phone duration results of this study, which are based on a corrected and extended version of the PhonDatII corpus, are in correspondence with earlier research. Consequently, the actual size of this spoken language resource seems to be sufficient for generalization of results on the segmental level.
Bibliographic reference. Pfitzinger, Hartmut R. (2002): "10 years of phondat-II: a reassessment", In ICSLP-2002, 369-372.