It doesn't detect them as multiple frequencies until later, it just samples the incoming sounds and creates a digital version of the analog measurements. from the mic.

There's a thing called 'digital signal processing' which then can then analyze the digital signal and break it up into it's composite frequencies/magnitudes, average out the noise. The way the processing is handled will determine the quality of the end result.

Some cool things to learn about would in Nyquist Theorem, Fast Fourier Analysis, and Digital Signal Processing (FFT is part of this but deserves it's own mention).