Skip to content

Hearing and perception

A sound pressure level says how much sound there is; this section is about what a listener makes of it. Its three subsections answer three different questions. Psychoacoustics quantifies sensations: how loud a sound is perceived to be, how sharp, rough or tonal it is, and how those sensations combine into annoyance. Speech asks how well spoken words survive the trip from a talker to a listener, through a room, a sound system or background noise. And hearing and exposure covers the ear itself: where the hearing threshold sits across age and population, how noise permanently shifts it, and how a working day’s exposure is measured and reported.

The three build on each other in one direction: the psychoacoustic models consume calibrated signals or band spectra from the core analysis; the speech indices consume band levels and, in the SII’s case, the hearing thresholds that the hearing pages quantify; and the exposure metrics feed the hearing-damage model of ISO 1999.

A good entry point is Loudness: it introduces the perceptual scale (the sone) and the auditory models that most other metrics in this section reuse or extend. The derivations behind these methods — the critical-band and excitation-pattern models, the masking formulations and the modulation-transfer chain — are gathered on the perception theory page, which the individual guides cite clause by clause.

Almost every perceptual scale in this section is defined by a reference sound rather than by a physical unit, so the first thing to learn about each is its anchor: the sound that reads exactly 1.

QuantityUnitThe sound that reads 1Criterion?Page
Loudnesssone1 kHz tone at 40 dB SPL (also 40 phon)noneLoudness
Sharpnessacumcritical-band-wide noise at 1 kHz, 60 dB SPLnoneSound Quality Metrics
Roughnessasper1 kHz tone at 60 dB, fully modulated at 70 Hz0.2 asper (informative)Sound Quality Metrics
Fluctuation strengthvacilthe same carrier modulated at 4 Hz0.2 vacil (informative)Sound Quality Metrics
Tonalitytu_HMS1 kHz tone at 40 dB SPL0.4 tu_HMS (informative)Sound Quality Metrics
Tone audibilitydB— (a level difference above masking)ISO 1996-2 adjustmentTone audibility
STI0 to 1Annex F letters, U to A+Speech Transmission Index
SII0 to 1none standardisedSpeech Intelligibility Index
Threshold shiftdB HL— (a difference of two hearing levels)ISO 1999 statisticsNoise-induced hearing loss

Loudness, sharpness, roughness and fluctuation strength are ratio scales with no pass/fail line: twice the number means twice the sensation, so a 20-sone appliance is heard as about twice as loud as a 10-sone one, which is why appliance declarations set limits in sones rather than in decibels. The tonal metrics and the speech indices do carry criteria, which is why the tone pages end in a verdict and the loudness pages do not.

The three speech numbers all live in [0, 1] and are not the same number. An STI of 0.6 falls in Annex F band D, typical of a good lecture room; an SII of 0.6 means roughly 60 % of the importance-weighted speech spectrum is audible; and a STOI of 0.6 has no absolute meaning at all, because the mapping from index to words understood is fitted per listening-test corpus, so STOI is only ever read as a difference between two processors on the same material. Never substitute one for another in a specification, and when a requirement quotes a number, check which standard it belongs to before computing anything.

The perceptual sensations of sound: loudness and the metrics layered on it.

Two complementary indices of speech intelligibility, STI for the transmission channel and SII for the listening condition, plus the signal-based STOI and ESTOI measures.

The hearing threshold, what noise does to it, and how exposure is measured.

  • No listener is tested here, and no verdict about a person is issued. Every model on these pages predicts what a population, or a standard listener, would perceive from a calibrated signal: none of them runs an audiometric session, none diagnoses a hearing loss, and ISO 1999 explicitly declines to define a hearing handicap or a compensable fence — that line is national regulation, and nothing here applies one. Every prominence and audibility verdict is likewise the numeric criterion only: ECMA-418-1 also requires aural confirmation of a prominent tone, and that stays with you.

  • Everything is monaural. The binaural combinations of ECMA-418-2 are not implemented, so a two-channel recording is analysed one ear at a time, and nothing here models localisation, spatial release from masking or binaural loudness summation.

  • No listening test is replaced. STOI returns the correlation-based index and not a percentage of words understood, because that mapping is fitted per listening-test corpus; the SII returns an audibility fraction and not a score; and no page predicts annoyance in a community, which is a social-survey quantity rather than a psychoacoustic one — the community indicators are Environment and transport.

  • Finally, these models start from a calibrated signal or spectrum in pascals, because every one of them is level-dependent. Feeding them raw soundcard samples produces a number with an arbitrary reference, which is a different failure from a wrong answer: it looks plausible.

Every model here consumes a calibrated signal or a calibrated spectrum, so the calibration and weighting that produce one are in Signal analysis, and Build a sound level meter runs that chain end to end on one runnable page. The derivations are in Perception and hearing theory, from the equal-loudness contours to the modulation transfer function.

If you arrived here from a search and want the shape of the whole library, What do you need to measure? indexes it by the job and All guides lists every page with a line on each.