SIGNAL CONDITIONING TECHNIQUES FOR ROBUST SPEECH RECOGNITION

Acoustic mismatch encountered in various training and testing conditio ns of hidden Markov model (HMM) based systems often causes severe degr adation in speech recognition performance. For telephone based speech recognition tasks, acoustic mismatch can arise from various sources, s uch as variations in telephone handsets, ambient noises, and channel d istortions, This paper presents three techniques for blind channel equ alization, namely, cepstral mean subtraction (CMS), signal bias remova l (SBR) and hierarchical signal bias removal (HSBR), Experimental resu lts on various connected digits databases show a reduction in the digi t error rate by 16%, 21%, and 28% when employing CMS, SBR, and HSBR, r espectively. Our results also demonstrate that the HSBR technique outp erforms SBR and CMS on every sub-data collection and exhibits consiste nt improvements even for short utterances.