Title:
Speech signal processing based on concept of amplitude modulation

Speaker:
UNOKI, Masashi (JAIST)

Abstract:
Humans can easily listen to target sounds that they want to hear in real environments, such as those that are noisy and reverberant one. In addition, hearing abilities can be improved by using attention. However, it is very difficult for machines (i.e., computers) to do the same thing. Implementing auditory signal processing with the same functions as those of human hearing systems onto computers would enable us to accomplish human-like speech signal processing. Such a processing system would be highly suitable for a range of applications, such as speech recognition processing and hearing aids. Achieving this is the ultimate goal of in my laboratory.

The following research projects have been used in an auditory filterbank to process speech signals: a selective sound segregation model, a noise reduction model based on auditory scene analysis, a speech enhancement model based on the concept of the modulation transfer function, and a bone-conducted speech restoration model for improving speech intelligibility. We usually used the gammatone auditory filterbank as the first approximation of a nonlinear auditory filterbank in these projects. Our main purpose was to model the 'cocktail party effect' and to apply this model to solving challenging problems by developing our research projects into a nonlinear auditory filterbank and attain auditory signal processing.

In this talk, research and developments are briefly introduced. In particular, speech signal processing based on concept of amplitude modulation will be introduced.