> Your ears contain millions of tiny fine hairs of varying lengths which each vibrate in response to some set of frequencies, making them essentially analog Fou

> Your ears contain millions of tiny fine hairs of varying lengths which each vibrate in response to some set of frequencies, making them essentially analog Fourier Transform devices. And then, your brain then does something stupidly complicated to this set of clean inputs, so that you can instantly tell whose voice is whose in a multi-speaker environment, and so that you can detect the slightest tremor in somebody’s voice that might clue you in on their mental state as they say those words. We undergo decades of musical training so that we can train our brains to unwind all of this complicated processing and extract pure tones from this jumble of sound. i always wondered why we dont seem to have developed ML models that can do this yet. its not like the synthetic data is hard to generate, if data limitations are the excuse.