The final files are at the bottom of this page, go ahead and listen to the FDR recording. It demonstrates how the meaning and coherence is lost as the sampling iterations continue.

The code below will take a wav input file. It then can use parameters to randomly sample a specified amount of times, with the samples being a desired length.

For my project I decided to sample three different historical speeches. The speeches are the following:

Franklin Roosevelt's "First Inaugural Address"

Martin Luther King's "I Have a Dream"

John F. Kennedy's "First Inaugural Address"

I sampled relatively long samples the first time, enough that the speech is recognizable. Coherent meaning is still recognizable in the first batch of 3-5 samples for each speech. Here are the first samples along with their code:

I then took the first batch of samples and ran the code on them again, this time sampling from the samples themselves. I shortened the sampling length from the original code and changed how many samples were taken. Words are still recognizeable, but it is harder to figure out what the speech is really about. Here are the second batch of samples and their code:

I did the process again on the second batch of samples. I shortened the length even more this time. Coherence is quickly leaving the speech. The appearance of stuttering (which I talk about later) begins to show. Here are the samples and code:

I did a fourth and final iteration by sampling the third batch with even shorter sample lengths. Almost no meaning can be derived from these short, repetitive sounds. Audio and code:

I then combined all the files together for each speech, forming a longer audio file that shows the process in order. Each file below starts off with the original speech, immeadiately followed by the first sample and then each one after that.

FDR Combined "First Inaugural Address"

MLK Combined "I Have a Dream"

JFK Combined "First Inaugural Address"

For some reason that I do not understand each iteration gets a little bit louder. The effect was unintened; however, I find it interesting since as the coherence and meaning of the speech are lost, the chaos is supplemented with added volume. After each iteration of sampling, all following iterations only have that part of the speech to work with. Due to this, the process has the appearance of "zooming in" on a particular part of the speech. At each step of the iteration the focus becomes even narrower.

The code I wrote can be applied to any wav files, such as songs or speeches. I had more trouble and struggled more than I thought I would with the coding, and as such I was unable to do as much as originally intened for this final project. Despite the unforseen troubles I ran into, I had a tremendous amount of fun once certain "boiler plate" problems were figured out.