Title :
Reducing the effects of linear channel distortion on continuous speech recognition
Author :
Bates, R.A. ; Ostendorf, M.
Author_Institution :
Dept. of Electr. Comput. & Syst. Eng., Boston Univ., MA, USA
fDate :
9/1/1999 12:00:00 AM
Abstract :
Linear channel compensation in speech recognition typically involves estimating an additive shift in the cepstral domain. This paper explores both Bayesian and maximum likelihood techniques to transform either the features or the model parameters. Experiments on the Macrophone corpus show error rate reductions of up to 16% over cepstral mean subtraction for short utterances
Keywords :
Bayes methods; cepstral analysis; maximum likelihood estimation; speech recognition; telecommunication channels; Bayesian technique; Macrophone corpus; additive shift estimation; cepstral domain; cepstral mean subtraction; continuous speech recognition; error rate reductions; experiments; linear channel compensation; linear channel distortion; maximum likelihood technique; model parameters; short utterances; Cepstral analysis; Channel estimation; Collision mitigation; Distortion; Hidden Markov models; Maximum likelihood estimation; Signal to noise ratio; Speech recognition; Telephony; Vectors;
Journal_Title :
Speech and Audio Processing, IEEE Transactions on
