Audio signals have been reduced to their 70%, 40%, and 5% most prominent spectral amplitudes (spectral peaks) by means of an overmasking function. Applying the commonly used spreading function with slopes of -27 dB/Bark and -24 dB/Bark results in the so-called Irrelevance Threshold with approximately 70% of all spectral components. According to the psychoacoustic theory of simultaneous masking no spectral components below that threshold can be perceived in steady state sounds. Nevertheless, the difference between the original and the processed signals can be made audible by spectral substraction. In case of 60% and 95% reduction, the irrelevance threshold is exceeded and the peaks of the amplitude spectra are enhanced. The intelligibility of speech signals drops increasingly and musical signals retain the leading voice only. The difference between the original and the processed signal can be perceived clearly. Again, the difference signal can also be made audible. The results are discussed in terms of psychoacoustic figure/background discrimination.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.