TRAINING A ROBUST COMMAND RECOGNIZER WITH THE TECNOVOZ DATABASE

This paper describes the development of a command-based robust speech recognition system for the Portuguese language. Due to an efficient noise reduction algorithm the system can be operated in adverse noise environments such as in vehicles or factories. The acquisition of a Portuguese database in the scope on the Tecnovoz project is addressed in this paper. The paper also describes a new noise-robust front-end and some experiments regarding the best acoustic model to use for a command-based speech recognizer. Results with whole-word, monophone and triphone models are presented and discussed. 1