MULTILINGUAL NON-NATIVE SPEECH RECOGNITION

摘要

The present invention relates to a method for selecting via a speech input a list element from a list of elements, the method comprising the following steps: - recognizing a string of subword units for the speech input, - comparing the recognized string of subword units to the list of elements and generating a candidate list of best matching elements based on the comparison result, wherein for generating the candidate list of best matching elements a confusion ma-trix is used which contains matrix elements comprising information about the confusion probability of subword units of different languages (comprisin g subword units of different natural languages).