InpaintingRevisited

Supplemental material to the paper Audio Inpainting: Revisited and Reweighted. by Ondřej Mokrý and Pavel Rajmic. For the interactive web page with playable audio, see https://ondrejmokry.github.io/InpaintingRevisited/.

A short guide to the files contained in the repository is provided below. In case of any questions, feel free to contact the authors.

Folders:

• docs – Source code for the web page.
• figures – Due to space limitations, there are only selected examples of the results in the paper. To provide a comprehensive comparison of the method, supplemental results are presented here. All the present figures are generated with scripts and data available in the folder matlab/data.
• matlab – Implementation of all the inpainting methods.

The matlab folder contains the following material:

• data – The subfolder contains results of all the performed experiments, together with scripts that plot the used figures; both data (*.mat files) and the scripts (plot_*.m) are divided into subfolders based on the respective method.
• Janssen – Functions for audio inpainting using AR-based method by Janssen et al. [1].
• OMP – Functions for audio inpainting using the OMP algorithm [2, 3].
• PemoQ – Functions for evaluation of the perceived quality of restored audio [4]. This folder is not part of the package due to copyright issues. However, it needs to be included to reproduce the presented results.
• reweighted l1 relaxation – Implementation of the l1-relaxation-based methods presented in the paper and supporting functions.
• SPAIN – Functions for audio inpainting using the SPAIN algorithm [5].
• wavs – This folder is created when the script demo.m is executed. The script then saves the restored signals as WAV files to this folder, sorted by signals.
• demo.m – The user is asked to choose signal, gap parameters (number, length) and algorithms from the preset options; the script then performs audio with such settings (if necessary, the parameters of the methods need to be changed in the m-file).
• EBU_SQAM.mat – Collection of signals used for the evaluation [6].
• global_test_... – Implementation of all the tests performed for the experimental part of the paper, except for the overall evaluation. These 4 scripts are the source codes which generated the files in the subfolder data.
• inpainting_comparison_full.m, inpainting_comparison_half.m – Implementation of the methods used for overall evaluation; the scripta reproduce all the data used in the paper, thus it takes a substantial amount of time to run them.

All the scripts were tested in MATLAB R2019a and using LTFAT [7, 8] (version 2.3.1), which is not contained in this package.

For the plotting scripts to be functional, MATLAB version R2018b or higher is necessary, because the function sgtitle() is used multiple times.

Furthermore, to work correctly, the interpreter for legend and titles should be set to latex, use for example

set(groot, 'defaultLegendInterpreter','latex');

set(groot, 'defaultTextInterpreter','latex');

References:

[1] A. J. E. M. Janssen, R. N. J. Veldhuis, and L. B. Vries, “Adaptive interpolation of discrete-time signals that can be modeled as autoregressive processes,” IEEE Trans. Acoustics, Speech and Signal Processing, 1986.

[2] A. Adler, V. Emiya, M. Jafari, M. Elad, R. Gribonval, and M. Plumbley, “Audio Inpainting,” IEEE Trans. Audio, Speech, and Language Processing, 2012.

[3] T. Blumensath, “Sparsify toolbox,” online, URL: https://www.southampton.ac.uk/engineering/about/staff/tb1m08.page#software

[4] R. Huber and B. Kollmeier, “PEMO-Q—A new method for objective audio quality assessment using a model of auditory perception,” IEEE Trans. Audio Speech Language Processing, 2006.

[5] O. Mokry, P. Zaviska, P. Rajmic, and V. Vesely, “Introducing SPAIN (SParse Audio INpainter),” in 2019 27th European Signal Processing IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING 20 Conference (EUSIPCO). IEEE, 2019.

[6] EBU SQAM CD: Sound quality assessment material recordings for subjective tests. online. URL: https://tech.ebu.ch/publications/sqamcd. [Online]. Available: https://tech.ebu.ch/publications/sqamcd

[7] Z. Prusa, P. L. Sondergaard, N. Holighaus, C. Wiesmeyr, and P. Balazs, “The Large Time-Frequency Analysis Toolbox 2.0,” in Sound, Music, and Motion. Springer International Publishing, 2014.

[8] P. L. Sondergaard. (2013) LTFAT webpage. URL: http://ltfat.sourceforge.net