Entwicklung eines Dynamikprozessors basierend auf psychoakustischer Modellierung

Diploma Thesis

A dynamic processor´s asset is to reduce the dynamic range of music or speech steplessly in a preferably imperceptible way. To achieve this aim, general psychoacoustic aspects plus pros and cons of contemporary approaches are discussed.

The first stage of the proposed approach requires an enhanced modelling of the course of the loudness including temporal masking effects. For this purpose the efficient "Warped-FFT" in conjunction with an advanced version of Karjalainen's inner ear model ("neural feedback model") is used. After that the "Static Function" determines the difference between the actual loudness and the target loudness. The following control process is adaptively conducted at psychoacoustically opportune moments. To retain spectral balance even at considerable alterations in amplitude, an additional "Multiband"-Version is proposed. In this case, all further calculations are carried out in a loudness-weighted time-frequency-domain. The final audio-processing is done with the aid of a dynamic "Warped"-Filter. The simpler "Fullband"-alternative utilizes a single gain coefficient for this purpose.

The developed approach is implemented in MATLAB®. Finally an informal listening test is conducted to compare the performance of both versions with current commercial solutions.