← Back to Audio Tools

Not Just Another SNR Number

Speech Intelligibility Test

Two setups can measure identical raw signal-to-noise ratio while one stays perfectly clear and the other turns to mush — it depends on exactly where the noise sits relative to the frequencies that carry consonant clarity. This measures that directly: how much a specific, realistic noise type masks the 1–4kHz band your words actually depend on, not just overall loudness.

Test Phrase

"The quick brown fox jumps over the lazy dog near the busy market."

Read this at your normal call volume, once in quiet and once with the noise profile playing.

Result

Pick a noise type above, then click Run Test.

Clarity Retained
  • Verdict:

How to Interpret Your Results

What your clarity-retained percentage actually means

This measures how much extra energy your chosen noise type dumps into the 1–4kHz articulation band — where consonant information actually lives — compared to your voice alone in that same band. A high percentage means the noise barely competes with the sounds that carry word clarity.

Over 85%Highly intelligible. This noise type barely touches the frequencies that carry word clarity — listeners are unlikely to struggle.
60–85%Somewhat degraded. Listeners will need to work a little harder to catch every word. Reducing the noise at its source helps more than raising your voice.
Under 60%Significantly degraded. This noise sits almost directly on top of your consonant frequencies — real intelligibility risk on a call.

Test more than one noise profile — a fan, typing, and chatter sit in different frequency ranges, and the same room can score very differently depending on which one is actually present.

Benchmark Reference

Clarity Retained at a Glance

GradeClarity retainedListener experience
Clear> 85%Words come through cleanly despite the background noise.
Strained60% – 85%Understandable, but listeners work harder than they should.
Degraded< 60%Real risk of being misheard or asked to repeat yourself.

Specific to the noise profile you tested — rerun with a different one to see how your risk changes.

Step-by-Step

How to Run the Test

  1. Pick a noise profile above — chatter, typing, or fan/HVAC hum — matching what you actually deal with.
  2. Click "Run Test" and allow microphone access when prompted.
  3. Read the test phrase once in quiet, then again once the noise profile starts playing.
  4. Read your clarity-retained result, then try a different noise profile to compare.

Notes

Why frequency overlap matters more than volume

Most of the information that lets a listener tell "pat" from "cat" from "bat" lives in a fairly narrow band, roughly 1–4kHz. A noise source that sits mostly outside that band — a low rumbling fan, for instance — can be fairly loud overall while doing comparatively little damage to intelligibility. A quieter noise that sits directly inside that band, like typing or nearby chatter, can meaningfully degrade clarity despite a better-looking raw SNR number. This test measures the overlap directly instead of assuming loud always means worse.

Common Issues

Fixing Poor Intelligibility

Mute when not speakingThe single most effective fix for any noise type — removes the masking entirely between your turns to talk.
Move the noise source further awayDistance reduces noise level faster than it reduces your voice, since you are presumably closer to your own mic.
Use a directional microphoneCardioid and shotgun mics reject more off-axis sound than an omnidirectional capsule, directly reducing how much background noise reaches the capture in the first place.
Microphone access blockedClick the camera/mic icon in your browser's address bar to confirm this site is allowed, or check your OS's microphone privacy settings.

Common Questions

Frequently Asked Questions

How is this different from the SNR reading in the microphone test?
Signal-to-noise ratio compares overall loudness to overall noise floor. This instead measures how much a specific realistic noise type masks the exact 1-4kHz frequency band that carries consonant information — two setups can have identical raw SNR while one stays clear and the other turns to mush, depending on where the noise energy actually sits.
Why does noise type matter more than noise volume?
Masking depends on frequency overlap, not just loudness. A low-frequency fan hum can be fairly loud while barely touching the consonant band, while quieter chatter or typing sitting right in that band can meaningfully hurt intelligibility despite a lower measured volume.
Is this a real speech intelligibility score like STI or PESQ?
No, and it does not claim to be. Those are formal, standardized measurements requiring dedicated modeling this tool does not have. This is a genuine, direct measurement of articulation-band masking — useful and honest, but a browser-based estimate rather than a certified score.
Why did I get a different score reading the same phrase twice?
Small variations in your speaking volume, pace, and exact mic distance between takes all affect the reading — this is a real, expected amount of natural variation, not a flaw in the test. Run it a couple of times and look at the pattern rather than a single number.
What noise types does this actually test against?
Real synthesized profiles shaped to match common problem sounds — a low-passed fan hum, a bandpass-filtered typing clatter centered where keyboard noise actually sits, and general chatter — rather than one generic white-noise stand-in for every real-world distraction.