3.3 Speech in noise: HINT, QuickSIN, AI / SII

The audiogram and quiet-WRS are necessary clinical measurements, but they are poor predictors of how a patient actually functions in the world. Most real-world listening is in noise — restaurants, cars, classrooms, family dinners — and noise interacts with hearing loss in ways that quiet measurements don’t capture.

A patient with mild-to-moderate sensorineural loss may have 92% WRS in quiet but be unable to follow a conversation at a dinner party. The disparity is the SNR loss: the additional signal-to-noise ratio the patient needs, compared to a normal-hearing listener, to achieve the same speech-recognition performance. Two people with identical audiograms can have wildly different SNR losses.

This lesson covers the modern speech-in-noise tests audiologists actually use, the articulation index (AI) and speech intelligibility index (SII) that quantify how a given audiogram predicts speech audibility, and the SNR-loss concept that ties them to function.

Why quiet WRS underestimates real-world difficulty

Three reasons:

1. Noise masks the softest phonemes first. A background noise at 50 dB SPL completely masks any speech component below 50 dB SPL at its frequencies. For a normal hearer this just makes things slightly harder; for a hearing-impaired listener, whose threshold is already elevated, the threshold and the noise floor can converge on each other, leaving no margin.
2. Sensorineural loss broadens cochlear filtering. Damaged outer hair cells reduce the cochlea’s frequency-selectivity — the listening “channels” each cover a wider frequency range. Noise outside the speech band leaks into the speech band more readily. This is off-frequency masking: a high-frequency noise can mask a low-frequency speech sound in a patient whose cochlear filters have widened.
3. Sensorineural loss impairs temporal-fine-structure processing. The fine timing cues that let listeners segregate a target talker from competing sound rely on auditory-nerve phase-locking that is degraded in cochlear pathology. This produces especially severe difficulties with competing-talker scenarios (“cocktail party problem”) that quiet WRS cannot detect.

The HINT (Hearing in Noise Test)

The HINT was developed by Nilsson, Soli, and Sullivan at House Ear Institute in 1994. The test uses 250 simple sentences (e.g., “The boy fell from the window”) presented in spectrally-matched speech noise. The audiologist adapts the speech level up or down depending on whether the patient correctly repeats; the noise level stays fixed.

The endpoint is the SNR at which the patient correctly repeats 50% of the sentences — the speech-reception threshold in noise, abbreviated SRTn or HINT SRT. Normal-hearing listeners achieve this at about −2 to −3 dB SNR; a patient with even mild SNHL may require +3 dB or more. The difference between the patient’s SRTn and the normative −2 dB is the SNR loss.

The HINT is widely used for cochlear-implant candidacy, hearing-aid outcome measurement, and research. The HINT-in-noise SRT is the standard “real-world function” number when the audiogram alone won’t do.

The QuickSIN

The QuickSIN (Quick Speech-in-Noise test), developed by Etymotic Research in 2004, is a short clinical alternative to the HINT. It uses six sentences per list, each containing five key words; sentences are presented at decreasing SNRs (+25, +20, +15, +10, +5, 0 dB) in a fixed sequence. The patient’s score is the SNR loss directly: it counts the words missed, applies a correction, and produces a single number in dB.

Interpretation:

The QuickSIN takes about a minute per ear and produces directly actionable information about how aggressively to pursue directional microphones, FM systems, or other noise-management interventions in hearing-aid fitting.

Articulation index → speech intelligibility index

The audibility-based theoretical underpinning is the articulation index (AI) of French and Steinberg (1947), modernised as the speech intelligibility index (SII) in ANSI S3.5-1997.

The AI / SII is a single number between 0 and 1 that summarises how much of the speech information at a given SNR is audible to a given listener:

where the sum runs over frequency bands (typically 20 critical bands), $I_i$ is the importance weight of band $i$ for speech intelligibility (an empirical table of values summing to 1), and $A_i$ is the audibility in that band — the fraction of the LTASS in that band that is above both the noise floor and the listener’s threshold.

For normal-hearing listeners in quiet, $A_i = 1$ in every band, so SII = 1.0 and intelligibility is at ceiling (typically > 95%). Background noise that exceeds the LTASS reduces $A_i$ in the affected bands; elevated thresholds in any band also reduce $A_i$. The two effects combine: an SII of 0.5 typically predicts about 70% sentence intelligibility, 0.3 predicts about 50%, 0.1 predicts near-zero.

The SII is the theoretical underpinning of nearly every hearing-aid prescription algorithm (NAL-NL2, DSL v5). The fitting target is, roughly, the gain that maximises SII for a “typical” speech-in-noise environment given the patient’s audiogram. We develop this in Ch 8 — Real-ear measurement and verification.

Closing the chapter

That closes Chapter 3. The arc: speech audiometry adds functional measurements to the sensitivity measurements of the audiogram. The SRT confirms (and sometimes contradicts) the pure-tone average. The WRS quantifies discrimination ability at a comfortable level and flags retrocochlear pathology when rollover appears. The speech banana makes audibility concrete by overlaying phonemes on the threshold curve. Modern speech-in-noise tests (HINT, QuickSIN) extend the picture into the noisy environments patients actually inhabit, and the AI / SII formula provides the audibility-to-intelligibility bridge that hearing-aid prescription rules are built on.

Next chapter: Ch 4 — Tympanometry and the middle ear. We leave behavioural testing for the first objective test in the toolkit — measuring the middle ear’s acoustic impedance with a probe in the ear canal.