Use of a reliability coefficient in noise cancelling by neural net and weighted matching algorithms

Discusses the problems of efficacy estimation in noise cancellation by a neural net-the lateral inhibition net (LIN)-and the use of this information in weighting matching algorithms. Since the effect of noise on the speech signal is variable and the backpropagation training algorithm is essentially stochastic (most common patterns have more influence in the weight re-estimation process), it is reasonable to suppose that the LIN's efficacy depends on the input, and each noisy frame could be associated with a reliability coefficient that attempts to measure how reliable the result of the neural net processing is. Isolated word recognition experiments have shown that reliability weighting can result in a mean error rate reduction as high as 96, 80, 58 and 36% at SNRs of 12, 6, 3 and 0 dB, respectively, when the noise is white Gaussian.