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

E0551 Dct Iter 2200

E0551 Dct Iter 2400

E0551 Learned Iter 200

E0551 Learned Iter 400

E0551 Linear Iter 200

E0551 Linear Iter 400

E0551 Linear Iter 600

E0551 Linear Iter 800

E0551 Linear Iter 1000

E0551 Linear Iter 1200

E0551 Linear Iter 1400

E0551 Linear Iter 1600

E0551 Linear Iter 1800

E0551 Linear Iter 2000

E0551 Linear Iter 2200

E0551 Linear Iter 2400

E0551 Nearest Iter 200

E0551 Nearest Iter 400

E0551 Nearest Iter 600

E0551 Nearest Iter 800

Deep Audio Prior

John Vinyard