These samples are from an experiment (more here: https://github.com/JohnVinyard/experiments/tree/master/deep-audio-prior) adapting the denoising results from the "Deep Image Prior" paper (https://dmitryulyanov.github.io/deep_image_prior) to the domain of raw audio samples.

These are checkpoints from every 200 iterations for four audio samples:

- speech
- hip hop
- drums
- classical piano

and four upsampling methods:

- linear
- nearest-neighbor
- learned (transposed convolutions)
- dct

Deep Audio Prior

7dba3 Dct Iter 1600

7dba3 Dct Iter 1600

John Vinyard