Cochlear Implant Atlas
CI Atlas · Candidacy & Evaluation · Module 05

5Speech-in-noise & ecologically valid testing

Real listening happens in noise, and candidacy testing has had to follow. A word test in quiet flatters a good hearing-aid user up to the ceiling and crushes a poor one to the floor, telling you little about the difficulty either meets in a noisy room where the signal-to-noise ratio sits around zero. So the modern battery adds adaptive sentence-in-noise tests. But noise testing hands the examiner a dangerous amount of power: by choosing the signal-to-noise ratio, or by separating the speech and noise loudspeakers so a hearing aid's directional microphones can help, a clinician can engineer almost any score — pass or fail. That makes the choice of test condition an ethical question, and the answer is a single fixed protocol applied identically to every candidate. This module is about testing honestly.

TWhy test in noise

Speech tests in quiet have two blind spots. In a good aid user they ceiling — everyone scores near 100%, hiding real disability; in a poor one they floor. Neither captures the everyday struggle, because daily listening happens at signal-to-noise ratios of roughly 0–10 dB. Adding noise restores the spread and exposes the handicap a candidate actually lives with.

The examiner can engineer the verdict — why the protocol must be fixed

050100sentence-in-noise (%)candidacy criterion-5051015signal-to-noise ratio (dB)

Quiet-only word tests ceiling in good aid users and floor in poor ones, so candidacy needs noise — daily listening sits around 0–10 dB SNR. But two examiner choices can manufacture almost any result: drop the SNR toward −5 dB and nearly anyone fails; separate the speech and noise loudspeakers (0°/180°) and a hearing aid's directional microphones inflate the score beyond anything a reverberant real room delivers. Presenting speech and noise from the same 0° speaker removes that artefact. The remedy for “cherry-picking” is a fixed, standardised protocol applied identically to every candidate, every session. Schematic.

CThe adaptive battery

Several validated tests make up the noise battery. HINT presents sentences in speech-shaped noise, adapting the level to find the SNR for 50% correct;[1994] QuickSIN and BKB-SIN use four-talker babble, scored as the speech-to-babble ratio for 50% correct;[2004] and AzBio sentences (lists equated through a cochlear-implant simulation) are tested in quiet and at fixed SNRs such as +5 or +10 dB. The shift toward AzBio came because HINT ceilings in many candidates, falsely disqualifying genuinely disabled aid users.[2008]

CGeometry and the directional-mic trap

Where the loudspeakers sit changes the score. If speech and noise come from the same 0° loudspeaker, a hearing aid's directional microphones cannot separate them — an honest, hard condition. Place the noise at 180°, behind the listener, and directional processing plus spatial release dramatically inflate the score, overestimating the benefit available in a real reverberant room. Co-located presentation is the conservative, valid choice for candidacy.

CThe ethics of test choice

Because the examiner picks the SNR and the geometry, they can effectively choose the verdict: test AzBio at −5 dB SNR and almost anyone fails; test easy and almost anyone passes. It is professionally improper to add noise testing only for patients who pass in quiet, or to hunt for the condition that yields the desired answer — that is selective testing to reach a result. The safeguard is a fixed, standardised protocol (a defined battery, level, SNR and geometry) applied identically to every candidate, every session. A telling pattern to watch for: good words but poor sentences in quiet flags a temporal-processing or auditory-memory deficit an implant may not fix.

Ceiling effects — why candidacy moved from HINT to AzBio

% scoring 100% (ceiling)28%HINT0.7%AzBio

A test that everyone aces cannot identify candidates. On HINT sentences, about 28% of would-be candidates score a perfect 100% — wrongly looking too good to qualify — whereas on the harder AzBio sentences only about 0.7% do. That single statistic is why the field retired HINT from candidacy in favour of AzBio: a battery must spread genuinely disabled listeners out, not pile them against the ceiling. After Gifford and colleagues. Illustrative.

Case 11.5 · Passes in quiet, drowns in noise
An adult scores 78% on aided sentences in quiet but reports they cannot follow conversation in any restaurant. A trainee proposes to record only the quiet score for candidacy.

What is the right approach?

Self-assessment — Module 52 questions
Question 1 · Trainee

Why is speech-in-noise testing part of cochlear-implant candidacy?

Question 2 · Clinician

Why must the speech-in-noise protocol be fixed and applied identically to every candidate?

Tracked locally in your browser — see /progress for the dashboard.