Effect of phase-sensitive environment model and higher order VTS on noisy speech feature enhancement [speech recognition applications]

摘要: Model-based techniques for robust speech recognition often require the statistics of noisy speech. In this paper, we propose two modifications to obtain more accurate versions combined HMM (starting from a clean and noise model). Usually, phase difference between is neglected in acoustic environment model. However, show how phase-sensitive model can be efficiently integrated context multi-stream model-based feature enhancement gives rise covariance matrices Also, by expanding vector Taylor series up second order term, an improved mean obtained. Finally, explain front-end itself preprocessing training data. Recognition results on Aurora4 database illustrate effect robustness each these modifications.