converting the audio or speech signal into a frequency spectrum S(k);

applying a non-linear function to at least one of a signal level A(k) representing an amplitude function of S(k) by a transmitter device and a signal level B(k) representing an amplitude function of S(k) by a receiver device, wherein

the step of applying the non-linear function to signal level A(k) is performed before the transmission of the signal and the step of applying the non-linear function to signal level B(k) is performed after the reception of the signal,

said signal level A(k) and signal level B(k) are based respectively on discrete amplitudes a(k) at the transmitter and discrete amplitudes b(k) at the receiver for frequency k and neighboring frequencies in signal level A(k) and signal level B(k),

application of the non-linear function to A(k) and B(k) modifies noise in the audio or speech signal such that noise in the audio or speech signal follows the audio or speech signal in frequency dependence across the frequency spectrum S(k);

pre-accentuation of the audio signal at transmission to provide a long-term frequency spectrum parallel to that of reception noise,

de-accentuation of the audio signal at reception,

applying the non-linear function and pre-accentuation and de-accentuation steps results in the obtaining of a frequency spectrum of the noise;

wherein, the frequency spectrum, which is originally flat, is modified in such a way that at the frequencies at which the audio or speech signal is strong, the noise level is augmented and at the frequencies at which the audio or speech signal is weak, the noise level is reduced,

whereby the noise level is modified so that the noise level remains below the useful signal for practically all frequencies, even if total noise power is modified unpredictably.