A Baseline System for Hands-Free Speech Recognition in car environment

Robust hands-free interaction is required for a wide diffusion of automatic speech recognition in the car environment. Under the European projects SpeechDatCar and VODIS~II we collected an Italian database of in-car speech consisting of 600 sessions recorded under various noise and driving conditions. In this paper we describe some recognition experiments on two tasks devised on a portion of this database: connected digit sequences and isolated command words. Performance obtained with HMMs trained on real data collected in the car is compared with that achievable with a speech contamination approach, which aims at reproducing realistic data on the basis of a clean speech corpus. Recognition performance is also analyzed as a function of the different noisy conditions and of the consequent SNR at the far microphones. Finally, the effect of HMM adaptation is investigated in order to tune the recognizer on the conditions of the various session