EXPERIMENTS WITH A NON-LINEAR SPECTRAL SUBTRACTOR (NSS), HIDDEN MARKOV MODELS AND THE PROJECTION, FOR ROBUST SPEECH RECOGNITION IN CARS

Achieving reliable performance in speech recognition in the car is an important challenge especially in the context of mobile telephony applications where the user can access the telephone functions by voice. The break through such a technology is appealing, since the driver can concentrate completely and safely on his task while composing and conversing in a full handfree mode. This paper adresses the problem of speaker-dependent discrete utterance recognition in the car-noise environment and mismatch context (training phase made in silence, recognition in noise). Experimental results are reported. The performance of a HMM-based recogniser rise from 31% (no compensation) to 98% after speech enhancement. More than 2000 utterances recorded in various conditions of noise have been used to test the system. This is achieved by the use of robust training/recognition schemes and by preprocessing the noisy speech by a novel non-linear spectral subtraction technique.