ROBUST WORD RECOGNITION IN ADVERSE CAR EN VIRONMENTS

This paper presents a novel word recognition technique which allows ha nd-free dialing in adverse car environments. The algorithm is based on speaker dependent word spotting. The paper compared and investigated such methods as spectral subtraction, short time modified coherence, e ffects of multi-microphone, dynamic and accelerated features, weighted distance measures and multi-templates by word recognition experiments . The experiments which were carried out using real speech database ut tered in adverse car environments showed that the proposed method usin g the cepstrum weighted lifter and multi-templates achieves 9896 word recognition accuracy in 80km/h driving car environments.