Short-Term Statistical Learning Mitigates the Ill-Posed Problem of Sound Localization

Area of Science:

• Auditory perception
• Psychoacoustics
• Computational neuroscience

Background:

• Auditory inference relies on integrating spectral and spatial sound cues.
• Listener anatomy shapes spectral cues for sound localization, but source-specific spectra can interfere, creating ambiguity.

Purpose of the Study:

• To investigate if listeners can mitigate auditory inference problems by statistically learning source spectral shapes over short timescales.
• To determine if short-term adaptation of spectral priors enhances spatial decoding accuracy.

Main Methods:

• A free-field sound localization task was employed.
• Participants localized ripple-spectrum sounds under predictable (fixed spectra) and unpredictable (randomized spectra) conditions.

Main Results:

• Sound source spectral predictability significantly reduced large-scale localization errors, including front-back reversals and quadrant confusions.
• Error reduction of up to 5% was observed within minutes of exposure to predictable spectra.

Conclusions:

• Listeners adapt their spatial decoding strategies by exploiting spectral consistency over short stimulus histories.
• This demonstrates empirical evidence for short-term updating of spectral priors in auditory inference.
• The findings highlight the adaptive nature of the auditory system in resolving spatial ambiguity.