Missing data mask models with global frequency and temporal constraints

摘要: Missing data recognition has been developped in order to increase noise robustness automatic speech recognition. Many different factors, including the decoding process itself, shall be considered locate masks. In this work, we are considering Bayesian models of masks, where every spectral feature is classified as reliable or masked, and independent from rest signal. This classification strategy can produce unrelated small ``spots'', while experiments suggest that oracle unreliable features tend clustered into time-frequency blocks. We call undesired effect: ``checkerboard'' effect. paper, propose a new missing classifier integrates frequency temporal constraints reduce, avoid, The proposed evaluated on Aurora2 connected digit corpora. Integrating such leads significant improvements accuracy.