If you listen carefully (and read the subtitles), this computer-controlled piano speaks English words in the voice of a child (via ScienceTip):
It's an interesting work of art and a cool (although not exactly innovative) engineering project, but what's surprising to me is not that reproducing a voice with piano keys is possible, but that it does it so poorly that captions are required. Naively, I would have thought that decomposing any waveform into 88 frequencies (or maybe 52 - it looks like only the white keys are used) would reproduce the original sound with decent fidelity, but it seems that's not the case.
Compare this to the way that we represent colours on a computer screen: just 3 frequencies (R, G, and B) are sufficient to represent any real-world colour which is, in fact, a continuous spectrum of frequencies — just like sounds. I know our eyes have just three different types of colour receptors and I'm sure that has something to do with it. Still, colours seem more complex to sounds to me, so this kinda blows my mind.
That's all for this week. Enjoy your weekend!