The "digits" small-vocabulary task is important for many telephone-based
applications such as computer-assisted long-distance dialing or credit-card billing,
requires extremely high accuracy, and focuses research on acoustic-level processing.
On the other hand, in many other tasks a speaker-independent domain-specific
vocabulary (such as "collect call", "calling card", "operator", or "help") needs to be
recognized. For such tasks, a "general-purpose" (gp) recognizer that is capable of
recognizing all permissible phoneme strings in a language is required.
The more recent results obtained by the application of the CSLU Toolkit frame-
based hybrid HMM/ANN architecture on these recognition tasks for the Italian
language are described. This work is inserted in a project whose aim is to contribute to
the " Italianization " of the CSLU Toolkit and to support the dissemination of these
tools and technologies.
