Coding and Decoding Speech using a Biologically Inspired Coding System

Citation

Pahar M & Smith LS (2020) Coding and Decoding Speech using a Biologically Inspired Coding System. In: 2020 IEEE Symposium Series on Computational Intelligence (SSCI). 2020 IEEE Symposium Series on Computational Intelligence, Canberra, Australia, 01.12.2020-04.12.2020. Piscataway, NJ, USA: IEEE. https://doi.org/10.1109/ssci47803.2020.9308328

Abstract
A spike (event) based sound coding technique has been presented in this study where the spikes are similar to the spikes exhibited by type 1 fibers of the auditory nerve. This lossy coding technique has already been shown useful for inter-aural time difference based sound source direction finding. Here, we show that decoding and resynthesising this code can produce intelligible speech even using a small number of spike trains. We have used few composite techniques including speaker verification to assess the effectiveness of the coding technique on a large number of TIMIT sentences. This biologically inspired coding technique can provide suitable input for a spiking neural network, as well as maintaining the accurate time structure of sound.

Keywords
spike coding; spike decoding; speech coding; biological inspiration; x-vector; i-vector; speaker diarisation; speaker recognition

People

Professor Leslie Smith

Emeritus Professor, Computing Science