Second European Conference on Speech Communication and Technology

Genova, Italy
September 24-26, 1991


Isolated Word Adaptive Recognizer Based on Neural Networks

Bruno Apolloni (1), Francesco Pazienti (2), Vincenzo Trotta (1)

(1) Dipartimento di Scienze dell'Informazione, Universita di Milano, Milano, Italy
(2) ALCATEL FACE Research Centre, Pomezia, Italy

This paper describes a general approach for implementing an isolated word recognizer based on neural networks. The target is the realization of a machine able to expand its knowledge with semi-automatic procedures, following the requirements of the environment. The vocabulary of the recognizer could be initially limited to few words selected by the constructor, then new samples can be added at run time. The growth of the vocabulary can be due to unaspected utterances of the same word, within the purpose of a speaker independent small vocabulary, or to new words, within the purpose of a speaker dependent large vabulary. Whenever the recognition is dubious the machine answers "I don't know", and starts a sequence of procedures intended to modify the lay out of the system, in order to allocate a larger vocabulary by a small interaction with the user. Keywords: speach recognition; voice signal compression; neural networks; back-propagation; learning algorithms; Kohonen maps.

Full Paper

Bibliographic reference.  Apolloni, Bruno / Pazienti, Francesco / Trotta, Vincenzo (1991): "Isolated word adaptive recognizer based on neural networks", In EUROSPEECH-1991, 101-104.