Noise reduction using neural lateral inhibition for speech enhancement

Research output: Contribution to journalArticle

Abstract

Recurrent spiking neurons with lateral inhibition connection play a vital role in human’s brain functional abilities. In this paper, we propose a novel noise reduction method that is based on neuron rate coding and bio-inspired spiking neural network architecture. The excitatory-inhibitory topology in the network acts as the temporal characteristic synchrony and coincidence detector that removes uncorrelated noisy spikes. A LIF source encoder is introduced along with the network. The network uses generated binary Short-Time Fourier Transform (STFT) masks according to the rate of processed spike train, which is used to reconstruct the denoised speech signal. The technique is evaluated on noisy speech samples with 5 types of real-world additive noise with different noise strength.
LanguageEnglish
Number of pages6
Journal International Journal of Machine Learning and Computing
Publication statusAccepted/In press - 10 Oct 2019

Fingerprint

Speech enhancement
Noise abatement
Neurons
Additive noise
Network architecture
Masks
Brain
Fourier transforms
Topology
Detectors
Neural networks

Keywords

  • spiking
  • neural network
  • speech enhancement
  • noise reduction
  • lateral inhibition

Cite this

@article{ca1f0fd5f7d64270872511c95604989d,
title = "Noise reduction using neural lateral inhibition for speech enhancement",
abstract = "Recurrent spiking neurons with lateral inhibition connection play a vital role in human’s brain functional abilities. In this paper, we propose a novel noise reduction method that is based on neuron rate coding and bio-inspired spiking neural network architecture. The excitatory-inhibitory topology in the network acts as the temporal characteristic synchrony and coincidence detector that removes uncorrelated noisy spikes. A LIF source encoder is introduced along with the network. The network uses generated binary Short-Time Fourier Transform (STFT) masks according to the rate of processed spike train, which is used to reconstruct the denoised speech signal. The technique is evaluated on noisy speech samples with 5 types of real-world additive noise with different noise strength.",
keywords = "spiking, neural network, speech enhancement, noise reduction, lateral inhibition",
author = "Yannan Xing and Weijie Ke and {Di Caterina}, Gaetano and John Soraghan",
year = "2019",
month = "10",
day = "10",
language = "English",

}

TY - JOUR

T1 - Noise reduction using neural lateral inhibition for speech enhancement

AU - Xing, Yannan

AU - Ke, Weijie

AU - Di Caterina, Gaetano

AU - Soraghan, John

PY - 2019/10/10

Y1 - 2019/10/10

N2 - Recurrent spiking neurons with lateral inhibition connection play a vital role in human’s brain functional abilities. In this paper, we propose a novel noise reduction method that is based on neuron rate coding and bio-inspired spiking neural network architecture. The excitatory-inhibitory topology in the network acts as the temporal characteristic synchrony and coincidence detector that removes uncorrelated noisy spikes. A LIF source encoder is introduced along with the network. The network uses generated binary Short-Time Fourier Transform (STFT) masks according to the rate of processed spike train, which is used to reconstruct the denoised speech signal. The technique is evaluated on noisy speech samples with 5 types of real-world additive noise with different noise strength.

AB - Recurrent spiking neurons with lateral inhibition connection play a vital role in human’s brain functional abilities. In this paper, we propose a novel noise reduction method that is based on neuron rate coding and bio-inspired spiking neural network architecture. The excitatory-inhibitory topology in the network acts as the temporal characteristic synchrony and coincidence detector that removes uncorrelated noisy spikes. A LIF source encoder is introduced along with the network. The network uses generated binary Short-Time Fourier Transform (STFT) masks according to the rate of processed spike train, which is used to reconstruct the denoised speech signal. The technique is evaluated on noisy speech samples with 5 types of real-world additive noise with different noise strength.

KW - spiking

KW - neural network

KW - speech enhancement

KW - noise reduction

KW - lateral inhibition

UR - http://www.ijmlc.org/

M3 - Article

ER -