A Noise-robust Auditory Modelling Front End for Voiced Speech

Citation

Smith L (1997) A Noise-robust Auditory Modelling Front End for Voiced Speech. In: Gerstner W, Germond A, Hasler M & Nicoud J (eds.) Artificial Neural Networks — ICANN'97: 7th International Conference Lausanne, Switzerland, October 8–10, 1997 Proceeedings. Lecture Notes in Computer Science, 1327. Artificial Neural Networks — ICANN'97, Lausanne, Switzerland, 08.10.1997-10.10.1997. Berlin Heidelberg: Springer, pp. 97-102. http://link.springer.com/chapter/10.1007/BFb0020139#; https://doi.org/10.1007/BFb0020139

Abstract
A method for detecting and displaying voiced elements of speech using amplitude modulated pulses due to unresolved harmonics of the excitation frequency (fundamental) is presented. It uses an auditory model consisting of a gammatone filterbank (modelling the basilar membrane), simple rectification (modelling the organ of Corti inner hair cells), envelope bandpass filters (modelling some spiral ganglion neuron effects) and amplitude modulation detectors (modelling certain cell populations in the cochlear nucleus). We demonstrate that it can display a pattern of activity across the spectrum and across time that describes the energy distribution in voiced speech, and that this pattern degrades slowly in the presence of non-speech noise.