Spectrogram Inversion

Audio Style Transfer

This secttion a companion page for the paper:
Wyse, Lonce. (2017), Audio spectrogram representations for processing with Convolutional Neural Networks. Workshop on Deep Learning for Music May, 2017, Anchorage

Figure 1 from the paper is reproduce here:

and the sounds for each spectrogram:

Style (rooster crow)
Content (speech: "The rich and famous")
a) With trained weights and no image noise
b) With random weights and no image noise
c) With trained weights and image noise
d) With random weights and image noise

All sounds were rendered from magnitude spectrograms using inversion techniques described here

...