Sixth International Conference on Spoken Language Processing
The goal of this work is to demonstrate the quality of multilingual automatic segmentations using the German MAUS system in order to substitute costly manually segmented data by automatically segmented corpora. In this study we investigated the influence of language specific HMMs in a cross-language task namely the automatic segmentations of English, Irench and Japanese with HMMs trained on German acoustic data. Given the orthographic transcription of an utterance we were able to produce quite good segmentations with the "wrong" acoustic models which will be described in detail in the following sections. The reason for this can either be based on the bigger influence of intra-/inter-speaker variability compared to the "interlingual variability" or on universal coarticulation processes as discussed below.
Bibliographic reference. Beringer, Nicole / Schiel, Florian (2000): "The quality of multilingual automatic segmentation using German MAUS", In ICSLP-2000, vol.4, 728-731.