Predictive Coding and Hearing

From free energy to everyday listening

Friston's free-energy principle casts brains as systems minimising surprise (prediction error) about sensory states.^[1] Clark translated the idea into "predictive brains" for situated agents: perception is controlled hallucination corrected by precision-weighted error signals.^[2] Sedley and Heerema-McKenney reviewed how auditory pathways implement predictive coding — from cochlear processing to belt cortex.^[3]

You do not hear raw pressure; you hear your best model of sources in a scene, revised millisecond by millisecond. Bregman's auditory scene analysis supplies the grouping rules that predictions operate over — streams, not pixels.^[8]

Mismatch negativity — error signals you never notice

When a repeating tone violates pitch, duration, or location, auditory cortex emits mismatch negativity (MMN) — an ERP deflection reflecting automatic prediction error. Näätänen and Schröger established MMN as a workhorse measure of auditory memory traces and models.^[4][5] Newborns show streaming biases consistent with early predictive structure (Winkler et al.), suggesting prediction is not a late cortical luxury.^[6]

MMN amplitude scales with salience and uncertainty: deviants that matter for behaviour get precision-weighted upward — linking predictive coding to attention capture.^[10]

Music, pleasure, and expected surprise

Predictive coding is not anti-surprise — controlled surprise delights. Gold et al. showed musical reward prediction errors correlate with pleasure onset; the brain enjoys competent violations within learned style rules.^[7] Koelsch's emotion reviews map expectancy fulfilment and breach onto limbic reward circuits.^[11] Norman-Haignere et al. found harmonic structure tuned in neural populations — predictions specialised for periodic complex tones in music and speech.^[12]

Ambient wellbeing sound usually wants low error: rain should stay rain, brown noise should stay brown. Pleasure-focused music wants structured error; conflating the two explains many bad "focus playlists" with random chimes.

Habituation as precision down-weighting

Rankin's habituation synthesis describes decreased response to repeated innocuous stimuli — functionally, the brain lowers precision on predictable input.^[9] Sokolov's orienting response is the opposite pole: novel input gets high precision, triggering body-level capture.^[10]

Identical four-minute loops fail twice: first habituation lowers calming effect; then a seam or repeat boundary spikes error — you hear the file, not the forest. Slow environmental change keeps predictions valid without brittle repetition.

Pathology angles — tinnitus as persistent error?

Rauschecker et al. proposed limbic-auditory loops in tinnitus where phantom predictions fail to resolve — error without an external deviant to explain away.^[16] Predictive accounts are speculative clinically but explain why masking and enrichment sometimes help: they supply external structure that lowers precision on internal error. Not a cure claim — a model guiding research.

Sleep, tinnitus, and precision at night

Sleep reduces but does not abolish prediction updating. Unexpected sounds spike precision briefly, provoking micro-arousals — Stanchina's ICU masking study shows how external randomness interrupts sleep even when subjective wakefulness is brief.^[14] Tinnitus may reflect persistent high-precision internal error (Rauschecker et al.); masking supplies external structure to compete with phantom predictions.^[16]

Riedy's review notes some sleepers benefit from continuous noise; others experience increased error when the masker itself becomes unpredictable through looping.^[20] Choose scenes with stable rules and slow change — predictive comfort, not surprise choreography.

Clark's predictive brains framing: perception is controlled hallucination corrected by precision-weighted error signals.^[2] In hearing, that means you experience what you expect until evidence forces an update — a door slam is a forced update at 3am.^[19]

Sedley and Heerema-McKenney mapped predictive coding across auditory cortex hierarchies — higher levels predict phonemes and streams, lower levels predict fine timing.^[3] When any level surprises, error propagates up; chronic surprise equals fatigue, a common report in open offices and tinnitus clinics alike.^[16]

Sound Bubbles as low-error gardens

Design choices map onto predictive comfort:

• Continuous brown floors — stable statistics, low surprise after initial onset.^[14]
• Spatial bubbles — sources stay in consistent places, matching spatial predictions.^[8]
• TimeLine Drift/Tide — micro-variation without boundary spikes.^[9]
• No random notification-like transients — avoids precision spikes during sleep or focus.^[14]

Stanchina et al. showed continuous noise reduced arousal to ICU spikes — externally supplied predictability buffering erratic hospital errors.^[14] Riedy's sleep-noise review notes individual variation: some brains weight prediction error differently at night.^[20]

Alvarsson et al. linked natural soundscapes to faster stress recovery — natural scenes carry learnable statistics evolved over millennia.^[15]

What predictive coding does not mean

It does not validate "manifestation frequencies" or claim thoughts create cochlear output. It is a computational description of neural inference. Cherry's selective listening shows top-down goals reshape which errors matter — not unlimited mind control over hearing.^[13]

Practical listening

1. Prefer scenes with stable rules you can learn in one sitting.^[8]
2. Avoid loops with audible seams — they are pure prediction-error bombs.^[9]
3. Mask external chaos with a gentle floor before error accumulates.^[14]
4. For pleasure listening, choose music that earns its surprises; for sleep, earn none.^[7]
5. Protect ears; damaged periphery feeds noisy errors upstream.^[17][18]

Gold et al.'s musical pleasure work shows prediction error can be rewarding when stakes are aesthetic, not survival — explaining ASMR whispers and drop-build EDM. Wellbeing gardens should sit on the low-error side of that spectrum unless listeners explicitly want musical surprise.^[7]

Koelsch's emotion reviews tie expectancy to autonomic coupling — heart rate and breathing follow predictive structure in music, another reason tempo-stable ambient beds feel physically calming.^[11]

From lab to living room

Predictive coding began as computational neuroscience; consumer audio can borrow the design insight without claiming to "retrain" Bayesian brains. Keep scenes stable, avoid seam bombs, separate streams in space, and prefer brown/nature statistics evolved over millennia.^[8][15]

Musical pleasure via controlled prediction error (Gold et al.) belongs in intentional listening sessions — not sleep gardens.^[7] Match error budget to context: high for concerts, near-zero for 3am.^[20]

Näätänen and Schröger's MMN reviews remain the bridge between millisecond ERPs and user-experience choices — deviants are never "free."^[4][5]

Summary

Hearing is predictive: Friston and Clark's frameworks, MMN lab signatures, and Gold's musical surprise map the same error machinery with different stakes.^[1][2][7] Wellbeing audio should minimise error at night; art audio may choreograph it by consent.

Sound Bubbles targets low-error gardens — stable brown beds, spatial consistency, slow TimeLine change — so listeners stay comfortable past the first hour.^[14]

Clark's situated-agent framing: prediction error is bodily, not abstract — why slam sounds hit the gut before the spreadsheet.^[2]

Sedley and Heerema-McKenney tie predictive coding to clinical audiology tests — MMN is not only academic.^[3]

Friston's free-energy frame is not wellness jargon — it is a precise claim about why seams and sirens hurt.^[1]

Norman-Haignere harmonic tuning shows prediction is spectral as well as temporal — timbre always matters here too now.^[12]

Limits

Predictive coding remains debated in detail — precision mechanisms, layer counts, and clinical translation are active research. Use the model to choose kinder sound environments, not to replace audiology or mental-health care.^[19]