Some results on the development of a hands-free speech recognizer for car-environment

This paper describes some activities being conducted at IRST with the aim of developing a technology for hands-free speech recognition in car environment. This technology is based on Hidden Markov Models and is being developed and evaluated by using the car database collected in the European projects SpeechDatCar and VODIS-II. Preliminary experiments are based on the use of filtered clean speech corpora for HMM training and on the application of MLLR adaptation to further reduce the mismatch between training and testing conditions. Results are promising but show the difficulty of this task, even when exploiting some material collected in the test environment for HMM adaptation