A Fully Convolutional Neural Network for Speech Enhancement (2016-09-22T00:00:00.000000Z)

TL;DR

The proposed network, Redundant Convolutional Encoder Decoder (R-CED), demonstrates that a convolutional network can be 12 times smaller than a recurrent network and yet achieves better performance, which shows its applicability for an embedded system: the hearing aids.

Abstract

In hearing aids, the presence of babble noise degrades hearing intelligibility of human speech greatly. However, removing the babble without creating artifacts in human speech is a challenging task in a low SNR environment. Here, we sought to solve the problem by finding a `mapping' between noisy speech spectra and clean speech spectra via supervised learning. Specifically, we propose using fully Convolutional Neural Networks, which consist of lesser number of parameters than fully connected networks. The proposed network, Redundant Convolutional Encoder Decoder (R-CED), demonstrates that a convolutional network can be 12 times smaller than a recurrent network and yet achieves better performance, which shows its applicability for an embedded system: the hearing aids.

Authors

Se Rim Park

1 papers

Jinwon Lee

1 papers

TL;DR

Abstract

Authors

References31 items

Image Denoising Using Very Deep Fully Convolutional Encoder-Decoder Networks with Symmetric Skip Connections

Deep Residual Learning for Image Recognition

Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin

Complex recurrent neural networks for denoising speech signals

A Simple Way to Initialize Recurrent Networks of Rectified Linear Units

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Adam: A Method for Stochastic Optimization

Convolutional Neural Networks for Speech Recognition

Learning spectral mapping for speech dereverberation

Iterative Closed-Loop Phase-Aware Single-Channel Speech Enhancement

Why do people fitted with hearing aids not wear them?

Rectified Linear Units Improve Restricted Boltzmann Machines

Understanding the difficulty of training deep feedforward neural networks

A short-time objective intelligibility measure for time-frequency weighted noisy speech

Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion

Babble Noise: Modeling, Analysis, and Applications

Performance measurement in blind audio source separation

Perceptual evaluation of speech quality (PESQ)-a new method for speech quality assessment of telephone networks and codecs

Acoustic properties of phonemes in continuous speech for different speaking rate

Speech enhancement based on a priori signal to noise estimation

A signal subspace approach for speech enhancement

DARPA TIMIT:: acoustic-phonetic continuous speech corpus CD-ROM, NIST speech disc 1-1.1

Enhancement and bandwidth compression of noisy speech

Suppression of acoustic noise in speech using spectral subtraction

A Regression Approach to Speech Enhancement Based on Deep Neural Networks

Speech enhancement with weighted denoising auto-encoder

Freesound 2: An Improved Platform for Sharing Audio Clips

Speech Enhancement Using a-Minimum Mean-Square Error Short-Time Spectral Amplitude Estimator

Layer Configuration Number of Filters Filter Width CED

Conv) Conv

Table 1: Network Configurations for CNNs: CED vs. R-CED vs

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names