Auditory Distance Perception in Wave Field Synthesis: Pre-Attentive Neural Processing in a Cocktail-Party Environment
Author
Zawistowski, Emil
Term
4. Term
Education
Publication year
2026
Abstract
This thesis examines whether the human brain automatically detects changes in sound source distance in a complex multi-talker cocktail-party environment when the sound field is rendered using Wave Field Synthesis (WFS). A 60-loudspeaker WFS array was used to create physically realistic sound fields that preserved natural distance cues, including wavefront curvature, direct-to-reverberant energy ratio, and intensity gradients following the inverse square law. Twenty young adults with normal hearing completed a passive auditory oddball paradigm in which they watched a silent movie while speech from a target talker alternated between a near (2 m) and a far (5 m) distance, embedded in a four-talker masker; intensity differences were deliberately left uncontrolled to maintain ecological validity. Neural activity was recorded with 32-channel EEG, and analyses focused on event-related potentials (Mismatch Negativity, MMN, and P3a) and a leave-one-subject-out CSP+LDA classifier for single-trial decoding of distance conditions. Contrary to theoretical expectations, no robust group-level MMN was observed for either near or far deviants, there was no amplitude or latency asymmetry between looming (approaching) and receding sources, and the P3a component provided no consistent evidence of involuntary attentional capture; an isolated significant p-value was contradicted by its Bayes factor. A preregistered many-standards SSA control confirmed the absence of a genuine mismatch response, and single-trial decoding of distance did not exceed chance performance. These null findings are interpreted in relation to methodological constraints, including an electrically incompatible EEG power supply and the demanding four-talker masking scene, which may have disrupted pre-attentive processing. Although the study does not support the idea that auditory distance functions as a strong predictive prior under the present conditions, it offers an important empirical benchmark and a reusable methodological framework (validated WFS-based stimulus generation, high-precision trigger synchronization, and a fully documented EEG pipeline) for future research on auditory distance perception, and highlights the value of combining Bayesian inference with conventional significance testing in technically challenging, ecologically valid paradigms.
Dette speciale undersøger, om den menneskelige hjerne automatisk registrerer ændringer i lydkilders afstand i en kompleks cocktail-party situation, når lydfeltet gengives med Wave Field Synthesis (WFS). Med en 60-højttaler WFS-array skabes fysisk realistiske lydfelter, der bevarer naturlige afstandskendetegn som bølgefrontkrumning, forholdet mellem direkte og efterklangsrige lydkomponenter samt intensitetsforskelle efter invers-kvadratloven. Tyve unge voksne med normal hørelse deltog i et passivt auditivt oddball-eksperiment, hvor de så en stum film, mens talesignaler fra en målkilde skiftede mellem en nær (2 m) og en fjern (5 m) afstand og blev maskeret af fire samtidige talere; intensitetsforskelle blev bevidst bevaret for at sikre en økologisk realistisk scene. EEG blev optaget med 32 kanaler, og analysen fokuserede på event-relaterede potentialer (Mismatch Negativity, MMN, og P3a) samt en leave-one-subject-out CSP+LDA klassifikator til enkelttidsafkodning af afstandsforhold. På trods af teoretiske forventninger blev der ikke observeret nogen robust MMN på gruppeniveau for hverken nær- eller fjern-deviant, ingen asymmetri i amplituder eller latenser mellem looming (nærmerekommende) og fjernere kilder, og P3a-komponenten viste ingen konsistent evidens for ufrivillig opmærksomhedsfangst; en enkelt signifikant p-værdi blev modsagt af Bayes-faktoren. En forregistreret many-standards SSA-kontrol bekræftede fraværet af en ægte mismatch-respons, og enkelttidsklassifikationen af afstand var ikke bedre end chance. Resultaterne diskuteres i lyset af metodiske begrænsninger, herunder elektrisk inkompatibilitet i EEG-strømforsyningen og den krævende fire-taler-maskeringsscene, som kan have hæmmet præ-attentiv behandling. Selvom studiet ikke understøtter hypotesen om, at afstand fungerer som en stærk prædiktiv prior under de givne betingelser, leverer det et vigtigt empirisk referencepunkt og en genanvendelig metodisk ramme (valideret WFS-stimulusgenerering, præcis triggersynkronisering og en fuldt dokumenteret EEG-pipeline) for fremtidig forskning i auditiv afstandsopfattelse og illustrerer værdien af at kombinere bayesiansk inferens med traditionel signifikans-testning i teknisk krævende, økologisk valide paradigmer.
[This abstract has been generated with the help of AI directly from the project full text]
Keywords