Synthesis of Perceptually Plausible Multichannel Noise Signals Controlled by Real World Statistical Noise Properties

Introduction

This website accompanies the JAES paper titled Synthesis of Perceptually Plausible Multichannel Noise Signals Controlled by Real World Statistical Noise Properties and provides supplemental material such as audio examples of analyzed and synthesized signals. Also, the MATLAB code which was used to generate the signals and the paper’s figures is provided.

The page is optimized to be viewed with modern web browsers such as Mozilla Firefox or Google Chrome. Users of Microsoft Internet Explorer will not be able to listen to the audio examples. However, the download links next to the audio players work just fine.

MATLAB Code

Obtain the code by cloning the Git repository by calling

git clone git@github.com:JAAdrian/AES2017NoiseSynthesis.git

or download the latest version from the master branch as a zip file and refer to the README.md document for further information on usage. The code is licensed under BSD 3-Clause and, thus, be freely altered and used as stated in detail in the license text.

Citation

When you use the provided code in your own research, please be sure to cite the following publication linked with the tools:

@article{adrian2017synthesis,
    author = {Adrian, Jens-Alrik and Gerkmann, Timo and van de Par, Steven
    and Bitzer, Joerg},
    title = {Synthesis of Perceptually Plausible Multichannel Noise Signals
    Controlled by Real World Statistical Noise Properties},
    year = {2017},
    pages = {914–928},
    volume = {65},
    journal = {J. Audio Eng. Soc},
    number = {11},
    url = {http://www.aes.org/e-lib/browse.cfm?elib=19360},
}

Audio Examples

In the following table, exemplary audio samples of analyzed and synthesized noise signals are presented. The audio samples include different types of noise which were considered in the associated paper. In the synthesis, a reduced parameter set was used as proposed in the paper.

Noise Type	Original	Synthesis
Cinema Optical Noise
Vinyl / Shellac
Cassette / Tape
Rain
Applause
Babble / Cafeteria Noise

Figure Generation

The MATLAB and data files which were used to create the more important figures in the paper are packed in this zip file. The file contains two folders. The folder ListeningTest contains subject data and evaluation scripts. The folder PlotScripts contains the scripts which produce the following plots:

Figure No.	Script Name
Fig. 4	plotFDLS.m
Fig. 5	plotParetos.m
Fig. 7	plotErrorDistribution.m
Fig. 8	plotListeningResults.m

Listening Test Results

Opinion scores obtained for five noise types with five conditions each. 12 subjects participated in this test, rating six signals for each condition (box). Pairs for which the null hypothesis of equal medians could not be rejected are marked with brackets. For multiple mutual equality the brackets span multiple groups.