World Statistics Day was yesterday, but we’re celebrating all week long! This celebration means acknowledging the impact statistics has on our world. Who is your favorite statistician? Share with us who they are and why they top your favorites list.
Choose Language Hide Translation Bar
Speaker art (and design of experiments) in the time of COVID

During these days of stay-at-home/work-at-home/education-at-home, many graphs, tables, models and statistics are reported and published about COVID. There are also plenty of stories, tales, accusations, reports and recommendations that have frightened, angered, inspired, and/or saddened. This blog post is none of those. It’s a simple article expressing gratitude for the opportunity to spend a little more time with my son (and work on a STEAM project), despite the circumstances from which it originated.

It all started several days ago when my son came to me with a few speakers he extracted from some speaker boxes purchased from ARC (a thrift store in the United States). He wanted to make paint splatter from them to make some art (speaker art is what he finally called it). But which speaker, he asked?

Well, I had to ask a few more questions. Where did you get this idea? How did you decide on the speakers presented to me? How are you going to make the speakers move? Which paint? What volume? Bass? Treble? What is the optimal combination of settings and features that will give us the optimal outcome? What is the outcome? What is optimal? I think some of you out there have a clue where I am going with this.

Let’s start with where he got this idea. It apparently started when he saw a YouTube video (during an educate-at-home-now-I-am-done-what-do-I-do-now kind of day). Maybe the video was about Martin Klimas’ art project on Sonic Sculptures? Maybe someone else trying something new?

I think I remember him saying it was some YouTuber who was messing around with different things you can do with speakers. That seems a little bit more like my son’s speed.

Second question: How did you decided on the speakers presented? Answer, the cones were waterproof or very water resistant, and they were smaller, more manageable. Such reason and thought! Here are pictures of four that he chose (speakers A – D):

Speaker ASpeaker ASpeaker BSpeaker BSpeaker CSpeaker CSpeaker DSpeaker D

Third question: How are you going to make the speakers move? Easy! With another purchase from ARC, a used Sanyo receiver! But there was a problem, my son noted. The filter capacitors (you will have to Google this one to learn more) were bulging and needed to be replaced (knowledge thanks to Mr. Carlson’s Lab, yet more YouTube educate-at-home time). So, after we replaced them with two new ones, we had a viable source for sound.

And the details go on, but what if we could apply design of experiments (DOE) to determine the optimal combination of features and settings? Brilliant! My son looked at me with a tilted head and said, “Say what?” I did a very glossed-over explanation that he understood and that seemed to satisfy him (he is 10 years old, by the way). Essentially, I said, “You want to know which speaker to use, right? Well, how are we going to find out using your receiver with all those different options?” And that was it. And here is where dad gets to show off JMP in all its glory (and show my son a little bit of what our company provides to our customers).

Enter the DOE: So long story made a little shorter, here is what we did:

  • Two continuous responses: Radius of spatter from center of cone in cm and spatter size
    • Medium spatter size was desired. This had its own caveats, neither of us were patient enough to measure that actual size of the droplets so we made a surrogate measure: None, Small, Mix between Small/Medium, Medium, Mix between Medium/Large, Large, Mix between Large/Extra Large, and Extra Large, which corresponded to a numerical scale: 0, 1, 1.5, 2, 2.5, 3, 3.5, and 4, respectively. So not perfect, but hey, it will work well enough. 2 (or medium) was considered the idea size which was a mix of small, medium and large droplets, but mostly medium.
    • Maximum radius was desired for maximum spread across the canvas (don’t forget, we are making art)
  • Three continuous factors: Volume (10,30), Amount of Bass (-10, 10), Amount of Treble (-10,10).
    • Volume: Dial read from 0 to 100. 0 would be no sound and not useful so 10 was easy to read and we could hear the sound. That was the lower bound. 30 was chosen as an upper bound because anything more could cause the cones (or speaker coils) to break on these low wattage speakers.
    • Amount of Bass and Treble was chosen as such since that was the min and max readings on the dials.
  • Three categorical factors: Speaker (four different ones), Loudness (on/off), and Hi-Filter (on/off)

We also wanted to test for interactions between all factors and any curvature in the continuous factors. Enter JMP Custom Design. We plugged in the above and originally got a 43-run design. We changed it to be 48 for good measure, and also because we could and we had time for it. Each run was estimated to take one to two minutes.

Here is the setup of the experiment (picture below):

  • Song: Radioactive by Imagine Dragons
  • Start and length of song to be used: Time Stamp of start 00:00 min:sec and duration 40 sec.
  • Volume of liquid: 10 mL (because it can consistently fit in all four speakers)
  • Liquid: Water, but later when optimal setting found will use watered down acrylic paint with similar density/viscosity
  • Connected speaker to Speaker Output A and to the right outputs (had to choose something since there were four possible outputs to connect to)
  • Played with Mono setting, not Stereo (we are using only one speaker at a time after all)

The SetupThe Setup

Example video snippet of a single run:

After collecting the data and analyzing it using the Model script created in the Table panel, we optimized for max radius and medium spatter size.


It looks like speaker A with Loudness on and Hi-Filter off using volume 30, with 10 for Bass and -10 for Treble (but 10 would have worked too and the results suggest that is was not that important of a factor). One quick note, the responses were zero inflated so I probably should have used a zero inflated distribution (like gamma found in Generalize Regression personality in Fit Model) for this, but Standard Least Squares was enough to get the basic idea (and it still worked out to be approximately the same answer).

With this in hand we made a few more paintings and voila! Art!

Speaker ArtSpeaker Art


After working on this, my son decided he wanted his own YouTube channel doing various other electronics-based projects. It really has got him thinking.

Thanks for reading!