Programming Etude #3: "Wekinate Your World"

by Marise van Zyl

Description of Etude 2:

The third programming etude for Music and AI, which can be viewed here, focused on using Wekinator. We had to come up with 3 different things to bring to life with interactive machine learning.

1. - Oraglif

I created a system that sees either a circle or a square and then plays either a sine wave or a square wave respectively. It is a completely original idea that is not inspired by anything at all.

See the ChucK code for 'oraglif' here.
See 'oraglif' in action below.

2. - The best boi

I made a camera monitor system for my house that will talk to Bruno (my dog) when I'm not at home so that he doesn't get lonely. It plays a random recording of my voice for him. I think he really likes it. Next step is to have it dispense an actual snack if he sits in front of the screen for more than a few seconds. Pavlov, but make it AI.

See the ChucK code for 'bestboi' here.
See 'bestboi' in action below.

3. - The worst zoom ever

For this part, I worked with Celeste and we made a networked system where you can control the other person's sounds with your camera input. It was tricky setting up, but once we got OSC working between the laptops, we were able to map sounds to the inputs and play around with different classes. It was super fun and I'm excited to play around more with networking.

See the ChucK code for 'the worst zoom ever' here.
'The Worst Zoom Ever' video coming soon.

Reflection:

I really really enjoyed Wekinator. Initially, I was very intimidated by it, but once I actually just got it working on one of the examples (it was actually very simple to set up) I REALLY enjoyed training models and making fun little things. I tried to make systems that are less chaotic for the first two examples, but I'm excited to make more expressive instruments, specifically moving towards final project thinking. One thing that I struggled with a bit, was fine-tuning my models. Sometimes, specifically with the shapes example, it couldn't distinguish between the circle and the square and I wasn't sure what to do to make it better. I figured I could just give more examples, but it would have been nice to know what kind of training would make it better. Are the examples too similar, should I provide more varied training where I'm holding the papers differently, should I adjust the lighting? It felt like the training/recording, and then classifying part was somewhat black-box-y in that I didn't know why it sometimes said square instead of circle, etc. I was very fun to be able to record the inputs though and see how they affected the model and figuring it out through trial and error.

Acknowledgements

Ge, YikAI, ChAI, Celeste, Bruno, Spencer Salazar (Auraglyph)