Using a typical human subject for binaural recording

Previousinvestigations haveshownthatuseofnon-individual binauralrecordingsresult ina substantially reducedlocalizationperformancewhencomparedtoreal-lifelistening andlisteningtoindividual binaural recordings. Thedeteriorations are mostpronounced in the medianplane, wherethe two earsreceivethe samedirectsound.Thepresent studycompriseda localizationexperiment, in which20 subjectslistenedto binaural recordingsfromthe ears of 30 humans.The "recordinghead" whichresultedin the best performancefor the groupof subjectswas selectedas "typical"and used in furtherexperiments.The number of median plane errors was 36.3%, when the non-individualrecordingsoriginatedfrom a randomsubject,and this figure was reducedto 21.2% withrecordingsfromthe typicalsubject.Thisvalueis not far from thereal lifevalueof 15.5%,althoughstillsignificantly higher.Front/backconfusionsin the horizontalplane werealmosteliminatedwithrecordingsfromthe typicalsubject, whereasthese werenumerouswithrecordingsfroma randomsubject.Theseresults wereobtainedwithindividual headphoneequalization. A marginal, althoughstatistically significantincreaseof errorswasseen, whena suitablecommonequalizationwas used.