Watching TinselKoala’s (TKLabs’s) transmission-line model demonstration, which uses an analog delay line made of inductors and capacitors to set up a standing wave that varies by frequency, with LEDs connected to show where the antinodes were, it occurred to me that perhaps you could use this approach for analog detection of voice frequency spectrum.
He’s using the interference patterns of the reflection in the unterminated analog delay line to get different patterns of lights on the LEDs. This is purely a linear phenomenon, so the pattern of standing voltages that will show up for signals at different frequencies is the sum of the voltages for the individual signals.
I think that you can get signals to bounce back and forth in the transmission line many times by using a large resistor on the input. Only harmonic frequencies will produce recognizable patterns of nodes and antinodes, I think? I don’t know how well this will work to separate harmonic frequencies from inharmonic ones.
Anyway, can you use these standing-wave shapes to detect a set of frequency buckets spread across several octaves? I think so. And you don’t need, necessarily, a compact analog delay line made of capacitors and inductors. You may be able to use AM modulation to upconvert a waveform from audio frequencies up to frequencies high enough that you can fit them into a short piece of coax, although perhaps they won’t be resonant frequencies.