LEARNING FRONT-END SPEECH RECOGNITION PARAMETERS WITHIN NEURAL NETWORK TRAINING

摘要

Techniques for learning front-end speech recognition parameters as part of training a neural network classifier include obtaining an input speech signal, and applying front-end speech recognition parameters to extract features from the input speech signal. The extracted features may be fed through a neural network to obtain an output classification for the input speech signal, and an error measure may be computed for the output classification through comparison of the output classification with a known target classification. Back propagation may be applied to adjust one or more of the front-end parameters as one or more layers of the neural network, based on the error measure.

主权项

1. A method comprising:
obtaining a power spectrum of an input frame of a speech signal; passing the power spectrum through a filter bank comprising a plurality of weights to create a filter bank output, each weight of the plurality of weights operating on a subset of frequency components of the power spectrum; processing the filter bank output to generate a set of features of the input frame; feeding the generated features through a neural network to obtain an output to classification; comparing the output classification to a target classification to compute an error measure; and applying back propagation to adjust the plurality of weights as a layer of the neural network based on the error measure.