Auditory Perception: Pitch, Timbre, and Scene Analysis Explained

Natural sounds often contain frequencies that are integer multiples of a fundamental frequency (f0). The brain interprets f0 as pitch, even when the fundamental itself is absent—a phenomenon known as the “missing fundamental.” Low‑numbered, resolved harmonics generate distinct peaks and valleys in the cochlear excitation pattern, allowing listeners to discriminate pitch more accurately than with high‑numbered, unresolved harmonics that overlap within the same cochlear filters. Neural tonotopy preserves this spectral detail, and machine‑learning models trained on speech and music develop human‑like pitch discrimination, suggesting that exposure to natural “auditory diet” shapes this ability.

Timbre and Loudness

Timbre describes the quality of a sound that remains after removing pitch and loudness. It depends on the relative amplitudes of overtones and on the temporal envelope, distinguishing, for example, a plucked string from a bowed one. Loudness follows a power law: perceived loudness grows roughly with intensity raised to the 0.3 power, so a 10 dB increase roughly doubles the sensation of loudness. Individual auditory‑nerve fibers possess a narrow dynamic range of about 25–30 dB, creating a “dynamic range problem” between the wide range of intensities humans can discriminate and the limited range of each fiber.

Auditory Scene Analysis

The auditory system faces an ill‑posed inverse problem: it must infer multiple sound sources from a single mixed waveform. Real‑world sounds occupy only a tiny fraction of the possible acoustic space, making statistical inference feasible. Perception operates as unconscious Bayesian inference, combining prior knowledge about typical sound structures with the likelihood of the current sensory data. Grouping cues such as common onset/offset timing and harmonicity serve as internalized statistical regularities that the brain uses to segregate a mixture into distinct sources—an essential step in solving the cocktail‑party problem.

Levels of Analysis

Understanding perception benefits from Marr’s three levels. The computational level asks what problem is being solved and under what constraints—here, extracting source identities from a mixture. The algorithmic level specifies the procedures, such as Bayesian updating and harmonic resolution, that the brain might employ. The implementation level concerns the neural circuitry that realizes these algorithms, including tonotopic maps and dynamic range adaptations in the auditory nerve.