How does this work?

[Jerchongkong]

I found your repository and I find it interesting, I just do not know how to make it work.

I wanted to test it with some mp3 to see how it works, I'm supposed to run "main.py" and I have to create a folder called "eval_rawdata" and put the mp3s there. its supposed to work but the generated wav files has the same quality than the mp3 seeing it with a spectrometer.

This is correct or there is another way to do it.

I think this project is abandoned :v

[flysofast]

Hi, thank you for your attention.
Sorry it was our attempt to adopt the idea of Image super resolution to see if it works with Audio as well (the network architecture is dead simple, but works well with Image SR though). At the early stage of my implementation the network actually produced much worse output than the input, so please give some credits to the network even though it looks like it doesn't do anything at all lol.
I have promised to myself to work more on that, like, 2 months ago. It’s summer laziness you know haha.
I have a few ideas on how to improve it:

• Use raw audio features instead of the STFT. The network has to handle the phase information as well as the magnitude, i.e. it should work with complex numbers.
• Bottleneck network may work better than the current network. Everybody says so.
• Log mel may not be the ideal feature since it omits phase information. However in my current implementation, I used the phase of the original low quality file to be that of output anyway. So log mel feature is also worth a try.

[LXP-Never]

you should try conv1D but not conv2D