Any non-linear component in the recording chain may introduce non-linear distortions. The characteristic of this kind of distortion is the creation of new frequency components in the spectrum, which may be non-harmonic in the worst case. Depending on the generation process of the new frequencies, harmonic distortions, intermodulation distortions and difference-frequency distortions are to be differentiated.
A system component shows a non-linear behaviour if the output amplitude is not proportional to the amplitude of the input signal. In other words, the output-to-input amplitude function is not a straight line but an arbitrary curve. Typical non-linear devices are limiters or noise gates. Every overload (i.e. when the signal exceeds the maximum dynamic range), including amplitude clipping and frequency aliasing, produces non-linear distortions. The resulting degraded speech signals are judged to be unclear, rough or croaky.
Due to the manifold factors influencing the detectability of non-linear distortions by human listeners (for instance loudness level and frequency components of the distorted signal), values for just noticeable distortions cannot be provided. However, the degree of intermodulation and difference-frequency distortions indicate the degree of degradation due to a non-linearity . Accordingly, for professional applications with high fidelity the intermodulation-distortion factor should be less than , since intermodulation distortion may already lead to audible effects.