Statistics for Adapting Large-Scale Speaker-Independent Automatic Speech Recognition to Dysarthric Speech