I had a lot of fun experimenting with the subroutines and Python Turtle methods yesterday but wanted to push it a little further and find out if I could make use of a new Python library to help create automated art.

Somehow I’ve never built a program that utilises and analyses audio before, so challenged myself to find out more about libraries such as PyAudio and Wave this afternoon. My daughter was practising piano in the other room so it gave me a push to integrate live audio into my solution, rather than rely on pre-recorded wav files.

I learned about numpy a little this afternoon too. I hadn’t realised it had functions to extract the frequency from an audio block (FFT). The more I explore Python, the more I fall in love with it as a language!

Once I’d successfully extracted numeric frequencies from the 5 second wave file into a list I looped through them and attempted to place shapes on the Python Turtle screen to correlate with the current frequency. I decided on a simple X axis plot to begin with but then, as I realised the range between min and max frequencies usually exceeded 8000 I introduced a scale factor so they could be seen on the screen together and adjusted the Y axis so that each frequency appeared bottom to top in the order of analysis.

Screen Shot 2016-03-31 at 18.18.40

Quite nice, but there’s a lot of white space where the unused frequency range lies. Instead of removing this range from the visualisation (which, in retrospect, might have been a good idea) I decided to attempt to create ghosts of the circles fading out as they get further from the original position. This led me into colorsys and all sorts of bother, reminding me (eventually) not to mess with anything that returns a Tuple until I convert it back to a List first. Anyway, I removed that part of the code and put my arty effects on the back burner. You can see one example of the mess below. Ugh.

Screen Shot 2016-03-31 at 18.19.00

I decided to alter the colour of the background this time too. I think I’d like to use some audio analysis to decide on the colour range in a future version so that low audio frequencies create darker images and high frequencies create bright, bubblegum pop images.

Screen Shot 2016-03-31 at 18.06.42

The last thing I added to the program was the option to use pre-recorded audio WAV files instead of always recording 5 seconds of audio. This was very easy to add as I’d modularised the code as I went, so all that was needed was a few lines extra in the main program:

Screen Shot 2016-03-31 at 19.08.33

Trying out the program with a few WAV files from or playing a YouTube video in the background resulted in the following images:

uptown funk
uptown funk

Python files can be found at Github – Feel free to fork the code, leave comments below or just enjoy the images it generates!

Leave a Reply

Your email address will not be published. Required fields are marked *