A perceptually motivated feature extraction method based on mimicking the masking properties of the cochlea has been recently found to provide enhanced performance when applied to conventional speech recognition back-ends. On the other hand, the introduction of Deep Neural Network (DNN) based acoustic models has produced dramatic improvements in performance. In particular, we found that Deep Maxout Networks, a modification of DNNs' feed-forward architecture that uses a max-out activation function, provides enhanced robustness to environmental noise. In this paper, we present preliminary experiments on the combination of these two elements that already show how the DMN-based back-end is capable of taking advantage of these auditorily inspired features making the whole system more robust and also suggesting that human-like representations of speech keep playing an important role in DNN-based automatic speech recognition systems.
展开▼
