Auditory Scene Analysis: How the Brain Organizes Sound

Equal loudness contours, also called Fletcher‑Munson curves, map the intensity needed for different frequencies to be perceived as equally loud. Detection thresholds follow a U‑shaped pattern, with the greatest sensitivity in the mid‑frequency range. At low sound levels, mid‑frequencies dominate perception; as overall intensity rises, the curve flattens and low and high frequencies become more audible. This shift explains why a mix that sounds balanced at a quiet listening level can seem thin or harsh when turned up.

Auditory Scene Analysis

Auditory scene analysis (ASA) confronts an ill‑posed problem: the ear receives a single acoustic waveform but must infer multiple, distinct sound sources, as illustrated by the classic “cocktail‑party” scenario. The brain resolves this ambiguity by applying internalized statistical regularities—perceptual priors—through unconscious Bayesian inference. Illusions that force the listener to mis‑group sounds reveal the constraints and priors that guide this inference process.

Grouping Cues

Onset and Offset

Abrupt amplitude changes signal the emergence of a new source, while gradual changes are interpreted as a single source waxing or waning. The “old‑plus‑new” heuristic captures this: frequencies present before a change are heard as continuing, and newly introduced frequencies are perceived as separate events.

Comodulation

When noise bands share a common envelope (comodulation), detection thresholds for a tone embedded in the noise improve by roughly 10 dB. This comodulation masking release demonstrates that the auditory system groups channels with identical temporal envelopes, making the tone stand out.

Harmonicity

Frequencies that are integer multiples of a fundamental are automatically grouped as belonging to the same acoustic event. Slight mistuning—detectable at as little as 2 % deviation—causes the mistuned component to segregate into a distinct tone, highlighting the brain’s sensitivity to harmonic relationships.

Repetition

Latent repetition within a signal acts as a powerful grouping cue. Even when mixtures appear inseparable, repeated patterns allow the auditory system to parse the components, effectively “locking onto” the recurring structure.

Spatial Cues

Binaural masking‑level differences show that interaural phase differences can improve detection thresholds by up to 20 dB. Spatial information thus aids the segregation of a target signal from background noise. The precedence effect further refines localization: reflections arriving later than about 5 ms are suppressed, allowing the direct sound to dominate perception. This suppression builds up over repeated presentations, enhancing spatial clarity in reverberant environments.

Streaming and Continuity

Stream Segregation

Alternating sequences of tones can be heard as a single melodic stream or as two separate streams, depending on frequency separation and presentation speed. When segregation occurs, listeners find it difficult to judge temporal relationships across streams, illustrating how perceptual organization shapes temporal judgments.

Continuity Effect

If a tone is briefly masked by noise, the brain often “fills in” the missing segment, perceiving the tone as continuous. This continuity effect reflects a strong prior that sounds persist unless there is clear evidence of interruption.

Phonemic Restoration

A speech analogue of the continuity effect occurs when noise bursts replace deleted phonemes. Listeners still perceive a complete sentence, demonstrating that the auditory system uses contextual priors to restore missing linguistic information.

Texture Inference

Stationary textures such as applause or rain are inferred to continue through long periods of masking. The brain’s expectation of stability for these textures leads to a perceptual inference that they persist even when they are physically absent.

Underlying Mechanisms

Bayesian perception formalizes the process: the brain selects the hypothesis about the world that maximizes posterior probability given the observed signal. This calculation combines a prior—knowledge of environmental regularities—with a likelihood derived from the acoustic evidence. The precedence effect, comodulation masking release, and harmonic grouping all exemplify how priors and likelihoods interact to produce coherent auditory perception.