Noise adaptive training using a vector taylor series approach for noise robust automatic speech recognition

In traditional methods for noise robust automatic speech recogni-tion, the acoustic models are typically trained using clean speech or using multi-condition data that is processed by the same feature en-hancement algorithm expected to be used in decoding. In this paper, we propose a noise adaptive training (NAT) algorithm that can be applied to all training data that normalizes the environmental distor-tion as part of the model training. In contrast to the feature enhance-ment methods, NAT estimates the underlying “pseudo-clean ” model parameters directly without relying on point estimates of the clean speech features as an intermediate step. The pseudo-clean model pa-rameters learned with NAT are later used with vector Taylor series (VTS) model adaptation for decoding noisy utterances at test time. Experiments performed on the Aurora 2 and Aurora 3 tasks, demon-strate that the proposed NAT method obtain relative improvements of 18.83 % and 32.02%, respectively, over VTS model adaptation. Index Terms — Noise adaptive training, model adaptation, ro-bust automatic speech recognition, vector Taylor series. 1