MARKOV MODEL-BASED PHONEME CLASS PARTITIONING FOR IMPROVED CONSTRAINED ITERATIVE SPEECH ENHANCEMENT

Research has shown that degrading acoustic background noise influences speech quality across phoneme classes in a nonuniform manner, This re sults in variable quality performance of many speech enhancement algor ithms in noisy environments. A phoneme classification procedure is pro posed which directs single-channel constrained speech enhancement. The procedure performs broad phoneme class partitioning of noisy speech f rames using a continuous mixture hidden Markov model recognizer in con junction with a perceptually motivated cost-based decision process. On ce noisy speech frames are identified, iterative speech enhancement ba sed on all-pole parameter estimation with inter- and intra-frame spect ral constraints is employed. The phoneme class-directed enhancement al gorithm is evaluated using TIMIT speech data and shown to result in su bstantial improvement in objective speech quality over a range of sign al-to-noise ratios and individual phoneme classes.