Acquisition of a Speech Database over the Telephone Network

The description of a speech database collected over the telephone by means of an automatic calling system is described. Particular care has been posed in the geographical coverage of the speakers. Both task-dependent and task-independent material has been collected, in order to build respectively a recognition system for connected digits and to provide a starting point for vocabulary-independent systems. All the speech material has been manually transcribed, including sponsaneous speech phenomena (breaths, coughs, laughs, hesitations) and acousticaly relevant noises