In this study we present different approaches to the recognition of non-natives. With a corpus in German spoken by speakers with 56 different first languages, the Strange Corpus, we perform recognition experiments with monolingual and multilingual recognizers. Among other, we compared two German recognizers, one that was trained in addition with non-native (Italian) speech and the other trained with German speakers only. We found that best performance is achieved with the recognizer trained with German including non-native speech, followed by a bilingual recognizer and an Italian recognizer trained with German and Italian natives.
Cite as: Uebler, U., Boros, M. (1999) Recognition of non-native German speech with multilingual recognizers. Proc. 6th European Conference on Speech Communication and Technology (Eurospeech 1999), 911-914, doi: 10.21437/Eurospeech.1999-222
@inproceedings{uebler99_eurospeech,
author={Ulla Uebler and Manuela Boros},
title={{Recognition of non-native German speech with multilingual recognizers}},
year=1999,
booktitle={Proc. 6th European Conference on Speech Communication and Technology (Eurospeech 1999)},
pages={911--914},
doi={10.21437/Eurospeech.1999-222}
}