Learning to Denoise Historical Music
This addresses the problem of audio restoration for archivists and music enthusiasts, but it is incremental as it applies existing neural network techniques to a specific domain.
The authors tackled the problem of denoising historical music recordings by proposing an audio-to-audio neural network model that processes spectrograms with a convolutional network trained on synthetic noisy data. Their results show the method effectively removes noise while preserving music quality, as validated by quantitative tests on synthetic data and qualitative human ratings on real recordings.
We propose an audio-to-audio neural network model that learns to denoise old music recordings. Our model internally converts its input into a time-frequency representation by means of a short-time Fourier transform (STFT), and processes the resulting complex spectrogram using a convolutional neural network. The network is trained with both reconstruction and adversarial objectives on a synthetic noisy music dataset, which is created by mixing clean music with real noise samples extracted from quiet segments of old recordings. We evaluate our method quantitatively on held-out test examples of the synthetic dataset, and qualitatively by human rating on samples of actual historical recordings. Our results show that the proposed method is effective in removing noise, while preserving the quality and details of the original music.