Music perception

According to Bayesian brain theories, perceiving music does not consist of passively receiving the sound signal: the brain continuously anticipates what is about to happen, then updates its internal model from the prediction error.

Predictions (priors): what the brain anticipates

The brain generates musical expectations from past experience, cultural exposure, context and the state of the subject. Several sources feed these priors: the expected musical structure (melody, harmony, rhythm); style, genres and cultural norms; musical memory and familiarity; goal or attention (active or background listening); emotional and bodily state. These musical filters are above all cultural: they are learned implicitly through long exposure to a musical corpus.

Sensory inputs: what the brain receives

The auditory system transmits sensory information and analyses it at several levels: timbres and sound qualities; pitches, intervals and chords; rhythm, tempo and micro-timing; dynamics (nuances, intensity); acoustic ambience and space. These ascending inputs correct or confirm the descending predictions.

Bayesian perceptual loop

Musical perception is an active inference, not a purely passive processing. The brain confronts its prediction (the expected) with the input (the heard). The prediction error is the signal that differs from the prediction: if the signal matches the expectation, the error is low and the model is confirmed; if it differs, the error rises and the model is updated. This error is at once the engine of attention, surprise and musical emotion.

Example: listening to a musical phrase

The plate illustrates the dynamic on a phrase. The brain anticipates the probable continuation; the real music is compared by the brain; in the case of a low error (fullness, coherence, stability), the model is confirmed; in the case of a moderate error (interest, engagement, frisson), it adjusts; in the case of a strong error (excessive surprise, confusion or rejection), the internal model is updated to better predict the continuation.

Optimal prediction-error zone

Applying the Wundt curve to music, the plate describes an optimum. Music that is too predictable produces boredom and disengagement. A slight error brings pleasure, relaxation and coherence. A moderate error provides interest, emotion, frisson and groove. An error that is too unpredictable generates noise, tension and rejection. Effective music navigates between predictability and surprise.

Neuroscientific data

Music listening activates prediction networks: the auditory cortex, the prefrontal cortex and the motor system (synchronisation, groove). The plate cites the electrophysiological mismatch response, the mismatch negativity (MMN), as an index of prediction error. Music also mobilises the motor systems, which links prediction and movement.

Result

Perceiving music is not a passive recording, but a dynamic dialogue between expectations and the real signal. Musical pleasure often emerges in a zone of moderate error: enough coherence to remain intelligible, enough surprise to remain alive. What we call music is in part an active construction of the brain.

Sources

• Huron (2006), Sweet Anticipation, musical expectations and emotions.
• Pearce and Wiggins (IDyOM models), statistical prediction of music.
• Salimpoor and colleagues (2011), dopaminergic response to musical pleasure.
• Vuust and colleagues (2022), prediction and rhythm.
• Friston, Clark, Hohwy, predictive coding and the Bayesian brain.