RRP - Audio in-painting

Abstract

In this contribution, we present a method to compensate for long duration data gaps in audio signals, in particular music. To achieve this task, a similarity graph is constructed, based on a short-time Fourier analysis of reliable signal segments, e.g. the uncorrupted remainder of the music piece, and the temporal regions adjacent to the unreliable section of the signal. A suitable candidate segment is then selected through an optimization scheme and smoothly inserted into the gap.

• Paper : Audio inpainting with similarity graphs
• Authors : Nathanael Perraudin, Nicki Holighaus, Piotr Majdak, Peter Balazs
• Date : June 2016
• ArXiv : http://arxiv.org/abs/1607.06667
• Online demo : https://lts2.epfl.ch/web-audio-inpainting/

In order to run this script you need to install the LTFAT and the GSPBox toolboxes. You can download them at:

• http://ltfat.github.io/
• https://lts2.epfl.ch/gsp/

Contents

A demo is availlable throught the file ai_demo_inpaint. The function ai_conf contains the default parameters.

Audio inpainting functions

• ai_audio_inpaint - Inpaint an audio signal
• ai_time_audio_graph - Create a graph of similarities from an audio signal
• ai_audio_inpaint_multiple_holes - Inpaint an audio signal with multiple holes
• ai_plot_graph - Plot the transitions graph
• ai_start - Start the audio inpainting box
• ai_compute_features - Compute a time-frequency features matrix
• ai_find_transitions - Find two optimal transitions

For help, bug reports, suggestions etc. please send email to nathanael (dot) perraudin (at) epfl (dot) ch