Annotation of a multichannel noisy speech corpus

This paper describes the activity of annotation of an Italian corpus of in-car speech material, with specific reference to the JavaSgram tool, developed with the purpose of annotating multichannel speech corpora. Some pre/post processing tools used with JavaSgram are briefly described together with a synthetic description of the annotation criteria which were adopted. The final objective is that of using the resulting corpus for training and testing a hands-free speech recognizer under development