Speech Enhancement based on Deep Convolutional Neural Network
Authors:
- 11 November 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Abstract
Speech enhancement is the process of treating noisy speech signals so as to improve human perception as well as improve system understanding of the signal. For speech signals with medium or high signal to noise ratio (SNR), the aim is to produce subjectively pragmatic signal, and for signals having low SNR the aim is to reduce the noise while still maintaining the intelligibility. Many noise reduction algorithms improve overall speech quality but little progress has been made to improve the overall speech intelligibility. This paper proposes a deep convolutional neural network (DCNN) speech enhancement method by enhancing loss function such as extended short time objective ineligibility (ESTOI) and mean square error (MSE). These loss functions are improved using Harris Hawks Optimization (HHO). The enhanced speech signal is acquired by separating the clean speech signal from the noisy speech signal. By using various predictive measure of objective speech intelligibility like short time objective intelligibility, source to artefact ratio (SAR), coherence speech intelligibility index (CSII) and source to distortion ratio (SDR), the efficacy of speech enhancement is calculated. The quality of the enhanced speech signal is assessed using the quality measure such as speech distortion (SD) and perceptual evaluation of speech quality (PESQ).Keywords
This publication has 21 references indexed in Scilit:
- Effects of Skip Connections in CNN-Based Architectures for Speech EnhancementJournal of Signal Processing Systems, 2020
- An enhanced productivity prediction model of active solar still using artificial neural network and Harris Hawks optimizerApplied Thermal Engineering, 2020
- A new perceptually weighted cost function in deep neural network based speech enhancement systemsHearing Balance and Communication, 2019
- NOISE-ADAPTIVE DEEP NEURAL NETWORK FOR SINGLE-CHANNEL SPEECH ENHANCEMENT2018
- Deep Neural Network Based Speech Separation Optimizing an Objective Estimator of Intelligibility for Low Latency Applications2018
- Phase-Sensitive Joint Learning Algorithms for Deep Learning-Based Speech EnhancementIEEE Signal Processing Letters, 2018
- Deep Recurrent Networks for Separation and Recognition of Single-Channel Speech in Nonstationary Background Audio2017
- Deep attractor network for single-microphone speaker separation2017
- Exploring Monaural Features for Classification-Based Speech SegregationIEEE Transactions on Audio, Speech, and Language Processing, 2012
- An analysis of a noise reduction neural network2003