High Performance Italian Continuous "Digit" recognition

The development of a speaker independent connected "digits" recognizer for Italian is described. The CSLU Speech Toolkit was used to develop and implement the system which is based on an hybrid ANN/HMM architecture. The recognizer is trained on contextdependent categories to account for coarticulatory variation. Various front-end processing was compared and, when the best features (MFCC with CMS + ?) were considered, there was a 98.68% word recognition accuracy (90.76% sentence recognition accuracy) on a test set of the FIELD continuous digits recognition task.