Subband{based Speech Recognition in Noisy Conditions: the Full Combination Approach Subband{based Speech Recognition in Noisy Conditions: the Full Combination Approach

In this report we investigate and compare di erent subband based Automatic Speech Recognition ASR approaches including an original approach referred to as the full combination approach based on an estimate of the noise weighted sum of posterior probabilities for all possible subband combinations We show that the proposed estimate is a good approximation of the ideal but often unpractical solution consisting in explicitly considering all possible subband subsets This approximation results in a nonlinear still simple and easy to implement combination function As opposed to other subband based approaches we believe that the proposed solution is more optimal mathematically correct and allows us to relax some of the subband independence assumptions Similarly to this full posterior combination approach which combines the subbands after independent processing a full feature combination approach is investigated in which all the possible subband features are orthogonalized and combined into a single feature vector before probability estimation The di erent approaches have been tested and compared on the Numbers database free format numbers with di erent levels of Noisex car noise This was done on the basis of two di erent acoustic features namely PLP and J RASTA PLP features and di erent weighting schemes Those experiments show that the full combination approximation yields very good estimates of the actual full combination posteriors and that both approaches yield very good recognition performance Acknowledgements The support of the OFES under the grant for the Speech Hearing and Recognition SPHEAR project OFES is gratefully acknowledged This paper bene ted from fruitfull discussions with my colleagues at IDIAP including Herv e Glottin Katrin Keller and Christopher Kermorvant as well as colleagues at other institutes such as Stephan Dupont at FPMs at Mons Belgium and Nikky Mirghafori and Brian Kingsbury at ICSI Berkeley USA