How to use it
The page has three tabs. Tone in noise plays a pulsing tone together with a band of noise. Press Play, change the noise and listen for the tone: it fades as the noise gets louder or wider, and it comes back when you move the noise away or cut a notch into it. Spread of masking plays a strong narrow noise at 1 kHz and a quiet probe tone that you can place below or above it. Measure is a short listening test: it finds how quiet a 1 kHz tone can be before noise of several widths hides it.
Headphones work best, but loudspeakers in a quiet room are fine. Start with the volume low.
Controls
- Volume and Stop: the slider sets the level of the whole page. Stop, or the Esc key, silences everything and pauses a measurement.
- Tone: on or off, frequency, level, and a pulse switch that makes a faint tone easier to notice.
- Noise: on or off, centre frequency, bandwidth from 10 Hz to 8 kHz, level per hertz, and a notch of adjustable width around the tone.
- Measure: answer with the two big buttons or the keys 1 and 2. Space or Enter continues after a pause.
How it works
When two sounds arrive together, the louder one can make the quieter one inaudible. This is simultaneous masking. The inner ear sorts sound by frequency, much like a row of overlapping band-pass filters. To detect a tone you listen through the filter centred on it, and only the noise that passes through the same filter gets in the way.
Harvey Fletcher showed this in 1940 with the band-widening experiment that the Measure tab repeats. The noise has the same power in every hertz and is centred on the tone. Its power inside a band B hertz wide is
so each doubling of the width adds 3 dB of noise. While the noise is narrower than the ear’s filter, the threshold of the tone rises with it. Once the noise is wider, the extra power falls outside the filter and the threshold stops rising. The width where the curve bends is the critical band: about 160 Hz near 1 kHz on the classic Bark scale. A later measure, the equivalent rectangular bandwidth of Glasberg and Moore (1990), is ERB = 24.7 × (4.37 F + 1) Hz with F in kHz, about 133 Hz at 1 kHz. Real auditory filters have sloping sides, so the bend is rounded.
The filters are not symmetrical. A masker hides sounds above its own frequency more than sounds below it, and more so when it is loud: the upward spread of masking. Masking also reaches across time. A loud sound can hide a quiet one that follows within roughly a tenth to a fifth of a second, and, far more weakly, one that comes a few milliseconds before it.
Lossy audio formats rely on all of this. A perceptual coder splits the sound into many frequency bands, uses a model of masking to estimate how much noise each band can hide at each moment, and stores each band only as precisely as needed to keep its rounding noise under that limit. Parts that would be masked anyway get few bits or none.
The test is a two-interval forced-choice task with a 2-down, 1-up staircase, which settles at the level you get right 70.7 % of the time. Steps are 4 dB, then 2 dB, and the threshold is the mean of the last six of eight reversals. Every burst of noise is built afresh from sine waves of equal strength and random phase, so its edges are sharp and its level per hertz is the same at every width.
Things to try
- Set the noise bandwidth to 10 Hz and widen it slowly. The tone fades until the orange block fills the shaded strip. After that, wider noise is louder but hides the tone no better.
- Slide the noise centre up to 3 kHz. The noise is as loud as before, yet the tone is clear again.
- Switch on the notch and widen the gap until the tone returns.
- Move the tone to 4 kHz and see how much wider the critical band is there.
- On the second tab, compare the two presets. The upper probe is much harder to hear than the lower one.
- Run the quick measurement twice. The saved runs show how much your thresholds move from one run to the next.
Frequently asked questions
Is this a hearing test?
No. It is not a medical test and says nothing about hearing loss. If your hearing drops suddenly, see a doctor the same day.
Why can I hear a tone that is weaker than the noise?
Only the noise inside one critical band competes with the tone. In wide noise the tone can be more than 10 dB below the total noise power and still be heard.
Why is my bend not at 160 Hz?
A home run has only a few dozen trials per point, so each threshold can move by 2 to 3 dB, and the experiment itself is not very sharp. Treat the bend as a rough estimate.
Does the volume change the result?
Hardly, as long as the noise is clearly audible and not loud. Tone and noise go up and down together, so their ratio stays the same.