Sound Generation with Neural Networks

Last Updated on Saturday, 17 July 2010 14:53 Written by Mehmet Ali ANIL Friday, 01 January 2010 23:12

One of the fascinating ideas that (and also some other friends) I try to to with is sound generation with Neural Networks which me and one of my friends do  some coding and realizations. One of the codes that we wrote was a Multilayered Perceptron, which we exposed to the input of an outer system, which was a Billings system, and the older output values of the original system. In the end the weights of the Multilayered Perceptron were changed due to the error composed by comparing the output of the original system, and the network.

For convenience and curious minds, the Billings System was defined as follows:

  That e(k) being random generated numbers, the neural network has its weights arranged so that it imitates the Billings system, after all, when we give the same white noise as we give the Billings system.

 Then I gave values of a sine wave to the input, e(k) and converted the result to sound: