AN ADVANCED FEATURE COMPENSATION METHOD EMPLOYING ACOUSTIC MODEL WITH PHONETICALLY CONSTRAINED STRUCTURE

This study proposes an effective model-based feature com-pensation method for robust speech recognition in back-ground noise conditions. In the proposed scheme, an acous-tic model with a phonetically constrained structure is em-ployed for the Parallel Combined Gaussian Mixture Model (PCGMM [1]) based feature compensation method. The structure of the acoustic model includes a collection of con-text independent phone models. A phonetically constrained prior probability is formulated by integrating transition prob-ability of phone models into the reconstruction procedure. Experimental results show that the PCGMM-based feature compensation employing the proposed phonetically con-strained structure of acoustic model consistently outperforms the case of employing the conventional Gaussian mixture model. This demonstrates that the proposed configuration of the acoustic model is effective at improving the intelligibil-ity of the speech reconstructed by the feature compensation method for speech recognition under diverse background noise conditions. Index Terms — feature compensation, PCGMM, acous-tic model, phonetically constrained structure, robust speech recognition 1