Any sound can be decomposed into a spectrum of frequency bins whose amplitudes change moment to moment — filtering just means damping selected bins.
X(f) = FFT(signal)
filtered(f) = X(f) if f < cutoff else attenuated
- Frequency bins — discrete frequency channels the audio signal is decomposed into via FFT.
- Bin connectivity — how much neighbouring frequency bins share energy (spectral leakage).
- Signal amplitude — loudness of the synthetic input waveform driving the spectrogram.
- Filter cutoff — frequency above which the signal is attenuated, reshaping the spectrum live.
This bin-by-bin frequency decomposition is exactly what powers noise-cancelling headphones and voice-assistant wake-word detection.