In a method for determining introduced into an audio signal information is a specific audio signal characteristic such. As estimated psychoacoustic masking threshold of the audio signal, which indicates the energy course of the inserted information. Then, the spectral representation of the audio signal is pre-processed on the basis of the characteristic. The now pre-processed audio signal is then subjected to a correlation to extract the information. Smoothing the widely varying energy of the audio signal leads to a simpler and better correlation and better extraction results. The smoothing can be achieved by suppressing tonal frequency bands or by inverse manipulating means of the psychoacoustic masking threshold, or by both measures. This also permits a corresponding modification in a device for introducing the information into the audio signal, such that no more perturbation is inserted into tonal frequency bands of the audio signal to avoid the risk of quality losses of the audio signal.