This paper presents the work done at Philips Research to extend our American-English large-vocabulary continuous speech recognition system to two new languages, French and German. The tasks on which the system is tested are very similar for each language: a speaker-independent, continuous-speech, national newspaper reading task with a high-quality microphone. However, an important factor for German is the extension of the recognition vocabulary from 20k to 64k words to achieve comparable out-of-vocabulary rates. A comparison between the different databases used to train the system for the three languages is made, results are accordingly interpreted and the characteristics of each language are pointed out.
Bibliographic reference. Dugast, Christian / Aubert, Xavier / Kneser, Reinhard (1995): "The Philips large-vocabulary recognition system for american English, French, and German", In EUROSPEECH-1995, 197-200.