A repository and webpage accompanying the paper Janssen 2.0: Audio Inpainting in the Time-frequency Domain.
The paper focuses on inpainting missing parts of an audio signal spectrogram. First, a recent successful approach based on an untrained neural network is revised and its several modifications are proposed, improving the signal-to-noise ratio of the restored audio. Second, the Janssen algorithm, the autoregression-based state-of-the-art for time-domain audio inpainting, is adapted for the time-frequency setting. This novel method, coined Janssen-TF, is compared to the neural network approach using both objective metrics and a subjective listening test, proving Janssen-TF to be superior in all the considered measures.
The paper compares a recent method abbreviated DPAI with the newly proposed Janssen-TF approach.
DPAI codes are not a part of this repository but are available here.
On the contrary, Matlab codes of our method are available in the Janssen-TF folder.
For reproducibility reasons, the codes are set to read the input (uncorrupted) audio files from the audio-originals folder,
while the spectrogram masks used in our experiments are read from the masks folder.
Note that there is several autoregression-based methods implemented in the Janssen-TF folder;
to exactly reproduce results from the paper, switch to ADMM, primal.
This provides the time-domain signal from line 7 of the algorithm in the paper, after the convergence is reached.
The Matlab codes for Janssen-TF use the LTFAT and Signal Processing Toolbox. To compute the perceptually-motivated evaluation, we have used the PEMO-Q package (version 1.4.1), which is not a part of this repository.