Decorrelation algorithms with application to artificial reverberation
Examensarbete för masterexamen
Decorrelation is a process which decreases the cross correlation between signals without destroying other qualities of those signals. In relation to audio, that means that the cross correlation of decorrelated signals is reduced but the audio is perceived as sounding the same. Decorrelation is known to reduce the effect of colouration, phantom sources and precedence effect in a multi-channel audio situation, but can also impart a greater spatial extent, diffuseness or headphone externalisation on sound . This thesis has two main parts. It first investigates and expands upon six different known decorrelation algorithms. The performance of these algorithms is analysed by various metrics, such as the extent to which they can decorrelate, impact upon the frequency content of an input signal and the group delay introduced. Three of the six decorrelation algorithms are selected as the best performing in these regards: IIR allpass cascade decorrelation, Sub-band decorrelation and White Noise decorrelation. The second part of the thesis is to use these three chosen algorithms to decorrelate the reverberant part of a Room Impulse Resonse (RIR) in order to simulate the diffuse sound field of sound in a room on a multi-loudspeaker setup inside a nonreverberant room. A listening test was performed, where participants were asked to rate sounds with different cross correlation or different decorrelation methods in terms of spaciousness and timbre. The conclusion drawn as a result of this listening test is that for the RIR with a longer reverberation time, a more decorrelated reverberant part of the audio does not result in a more spacious perception, but for the less reverberant RIR, it does. The Sub-band and White noise decorrelation algorithms produce the most perceptually spacious sound. The most decorrelated signals tested produce the largest difference in timbre when compared to less decorrelated signals, but it is important to note that all levels of decorrelation have a small to moderate difference in timbre when compared. Further investigation into this topic would most usefully be focused on the cause of an increase in low frequency content of audio for the least decorrelated audio when compared to less decorrelated. This was noted in informal conversation with most of the subjects that participated in the listening test, but not conclusively evidenced in the results of that listening test. It is suspected that this low frequency “booming” characteristic of the stimuli was percieved as spacious sounding, when it was not necessarily an intended part of any decorrelation algorithm.
decorrelation, correlation, room impulse response, listening test