Source: https://jmrplens.github.io/phonometry/perception/psychoacoustics/ # Psychoacoustics Psychoacoustics replaces the question "how many decibels?" with "what does the listener perceive?". Its base quantity is **loudness**: a perceptual magnitude in sones, computed by auditory models that account for the ear's filtering, masking and compression. On top of loudness sit the **sound quality** sensations that distinguish two equally loud sounds: sharpness (high-frequency emphasis), tonality (audible discrete tones), roughness (fast modulation) and fluctuation strength (slow modulation). And on top sits a combined **annoyance** metric that weighs loudness, sharpness, roughness and fluctuation strength into a single scalar. Every metric here is a magnitude fixed by a **reference sound** rather than by a physical unit, and knowing the anchor is what makes a number readable: 1 sone is a 1 kHz tone at 40 dB SPL, 1 acum a critical-band-wide noise at 1 kHz and 60 dB, 1 asper a 1 kHz carrier fully modulated at 70 Hz at 60 dB, 1 vacil the same carrier modulated at 4 Hz, and 1 tu_HMS a 1 kHz tone at 40 dB. They are tabulated together, beside the speech and hearing scales, under "Reading the numbers" on the [section overview](https://jmrplens.github.io/phonometry/index/). The two families of pages differ in purpose, and that difference decides what you can conclude. Loudness, sharpness, roughness and fluctuation strength are **open-ended magnitudes** for comparing designs: there is no pass mark, and the useful statement is always a comparison. The two tonal pages end in a **verdict against a criterion**, because they exist to justify a declaration or a penalty. ECMA-418-2 sits between the two: it attaches informative prominence criteria to its tonality (0.4 tu_HMS on a band), roughness (0.2 asper) and fluctuation strength (0.2 vacil_HMS), which is the closest thing to a pass mark in the magnitude family. All of them share one prerequisite: an absolutely calibrated signal in pascals, because every metric here is level-dependent. [Loudness](https://jmrplens.github.io/phonometry/perception/loudness/) is the foundation page: the Zwicker reference method of ISO 532-1 with its one-page fiche, together with the ISO 226:2023 equal-loudness contours that anchor the perceptual scale for pure tones. The newer model families, Moore-Glasberg per ISO 532-2/-3 and the Sottek Hearing Model of ECMA-418-2, continue in [Advanced Loudness](https://jmrplens.github.io/phonometry/perception/advanced-loudness/), which also carries the model-choice table. [Sound Quality Metrics](https://jmrplens.github.io/phonometry/perception/sound-quality/) adds sharpness per DIN 45692 and the ECMA-418-2 tonality, roughness and fluctuation strength that share the Sottek front-end. Tones in noise get two dedicated pages because two different questions are asked of them. [Prominent Discrete Tones (ECMA-418-1)](https://jmrplens.github.io/phonometry/perception/tone-prominence/) answers a product-noise question: is this tone *prominent* by the tone-to-noise and prominence-ratio criteria used in IT-equipment declarations? [Objective audibility of tones in noise (ISO/PAS 20065)](https://jmrplens.github.io/phonometry/perception/tone-audibility/) answers an environmental one: by how many decibels does the tone exceed its masking threshold, the audibility that feeds the tonal penalty of ISO 1996-2. [Psychoacoustic annoyance and fluctuation strength](https://jmrplens.github.io/phonometry/perception/psychoacoustic-annoyance/) closes the chain with the Fastl & Zwicker model, which combines loudness, sharpness, roughness and the slow-modulation sensation of fluctuation strength into a single annoyance value. Read it last: three of its four inputs come from the earlier pages, and it supplies the fourth, fluctuation strength, itself, in both the Fastl & Zwicker closed form and the Osses 2016 signal model. The ECMA-418-2 fluctuation strength on the Sound Quality page is a further, normative model of the same sensation, under a different unit name. ## Pages in this section - [Loudness](https://jmrplens.github.io/phonometry/perception/loudness/): the ISO 532-1 Zwicker loudness in sones, plus the ISO 226:2023 equal-loudness contours. - [Advanced Loudness (ISO 532-2/-3, ECMA-418-2)](https://jmrplens.github.io/phonometry/perception/advanced-loudness/): the Moore-Glasberg and Sottek loudness models and the model-choice table. - [Sound Quality Metrics](https://jmrplens.github.io/phonometry/perception/sound-quality/): sharpness (DIN 45692) and ECMA-418-2 tonality, roughness and fluctuation strength. - [Prominent Discrete Tones (ECMA-418-1)](https://jmrplens.github.io/phonometry/perception/tone-prominence/): tone-to-noise and prominence ratios with prominence verdicts. - [Objective audibility of tones in noise (ISO/PAS 20065)](https://jmrplens.github.io/phonometry/perception/tone-audibility/): the audibility of a tone above the masking threshold, feeding the ISO 1996-2 tonal adjustment. - [Psychoacoustic annoyance and fluctuation strength](https://jmrplens.github.io/phonometry/perception/psychoacoustic-annoyance/): the Fastl & Zwicker annoyance model and the fluctuation-strength models it consumes. ## What this section does not cover **Everything here is monaural.** The binaural combinations ECMA-418-2 defines for loudness, roughness and fluctuation strength are not implemented, so a stereo or binaural recording is analysed one channel at a time, and no model here accounts for localisation or for spatial release from masking. Two optional refinements are also left out: the entropy weighting of clause 7.1.6, which needs an external rotational-speed signal, and the small adjustment footnote 47 permits. **A verdict is never complete.** The `prominent` flag the tone-prominence functions return is the numeric criterion alone; ECMA-418-1 also requires aural confirmation and a lower-threshold-of-hearing screen, both of which stay with the caller. The tone-audibility module is weighting-agnostic and does **not** apply the A-weighting clause 5.3.2 requires, so A-weight the spectrum before passing it, and it takes an already-computed narrow-band spectrum rather than building one from a recording. Two documented deviations are worth knowing. ISO 532-3 prescribes resampling to 32 kHz before the running FFT; this implementation works at the native rate, a deviation that stays inside the standard's expanded uncertainty but that you should undo by resampling first if strict clause-by-clause conformance matters. And the Osses 2016 fluctuation-strength signal model is validated for amplitude-modulated stimuli only, with a documented floor — a steady 1 kHz tone reads about 0.09 vacil rather than 0. Finally, none of these metrics is a community response: annoyance here is a laboratory sensation computed from a signal, while the annoyance a neighbourhood reports is a social-survey quantity handled through the indicators of [Environment and transport](https://jmrplens.github.io/phonometry/environment/).