出版社：Academy & Industry Research Collaboration Center (AIRCC)
摘要：Single-channel speech intelligibility enhancement is much more difficult than multi-channelintelligibility enhancement. It has recently been reported that machine learning training-basedsingle-channel speech intelligibility enhancement algorithms perform better than traditionalalgorithms. In this paper, the performance of a deep neural network method using a multiresolutioncochlea-gram feature set recently proposed to perform single-channel speechintelligibility enhancement processing is evaluated. Various conditions such as differentspeakers for training and testing as well as different noise conditions are tested. Simulationsand objective test results show that the method performs better than another deep neuralnetworks setup recently proposed for the same task, and leads to a more robust convergencecompared to a recently proposed Gaussian mixture model approach.
