Perceptual Sine Sweep — 20 Hz–20 kHz

A perceptual sine sweep walks the full audible range the way hearing works — in equal bands, not equal hertz. Thirty-two ERB-spaced steps run from 20 Hz to 20 kHz with equal dwell per band, so the bass advances slowly and the treble moves quickly while pitch climbs evenly. Watch the live frequency readout, follow the progress bar, and run the sweep up and down. Everything is generated live in your browser, and nothing is uploaded.

How to use it

1. Set output level first, then forget it. Begin at a low, comfortable volume. The sweep holds one dBFS level throughout; the point is to hear system and ear behavior, not to chase loudness.
2. Choose the grid. 16 to 48 steps and 12 to 48 seconds, up or down. The default 32 steps over 24 seconds spends about 0.75 seconds on every hearing band.
3. Play upward and read the display. The live frequency value shows each step landing: roughly 20, 58, 102, 152, 210 Hz for the first steps, then thousands of hertz per step near the top.
4. Play downward. The same limits appear in reverse order, and adaptation usually makes one direction feel clearer. Note which steps ring, drop out, or vanish.
5. Follow up numerically. Where the sweep flags something, hold that exact pitch on the tone generator and stare at it with the spectrum analyzer before drawing conclusions.

Why steps follow hearing, not hertz

Human frequency selectivity is modeled by the equivalent rectangular bandwidth (ERB), the width of an auditory filter that passes the same energy as the cochlea’s response at that frequency. The standard Glasberg & Moore (1990) formula is ERB(f) = 24.7 × (4.37 × f/1000 + 1) hertz: about 27 Hz wide at 20 Hz, 133 Hz wide at 1 kHz, and 2,184 Hz wide at 20 kHz. Map frequency onto that scale with the ERB-number function G(f) = 21.4 × log10(4.37 × f/1000 + 1), and the audible range spans only about 41 bands: G(20 Hz) = 0.78 versus G(20 kHz) = 41.65. This tool divides that 40.9-band span into equal steps of 1.32 bands each, so every transition is the same perceptual distance even though the hertz distance varies by nearly three orders of magnitude.

The table shows actual grid points computed from the same formula the sweep engine uses.

Step	Frequency	Distance to next step
0	20 Hz	+38 Hz
1	58 Hz	+44 Hz
2	102 Hz	+50 Hz
3	152 Hz	+58 Hz
…	…	…
28	12,988 Hz	+2,015 Hz
29	15,003 Hz	+2,322 Hz
30	17,324 Hz	+2,676 Hz
31	20,000 Hz	end

A linear 32-step sweep over the same range would take 644 Hz leaps everywhere, giving the bass region a single grid point and the treble dozens it cannot distinguish. A logarithmic sweep moves at a constant frequency ratio per second, already bass-friendly, but the ERB grid goes one step further: it allocates time in proportion to how finely the cochlea actually resolves sound.

What the constant-level sweep reveals

The generated tone never changes amplitude in dBFS, yet your experience of it will not be flat, and that gap is itself information. Equal-loudness contours, measured from Fletcher & Munson’s 1933 work through the modern ISO 226:2003 revision, show the ear is most sensitive between roughly 2 and 5 kHz — thanks to ear-canal resonance and middle-ear transfer — and dramatically less sensitive at the extremes, with required SPL climbing as steeply as 24 dB per octave as tones descend toward 20 Hz near quiet levels. So a constant-dBFS sweep sounds like it swells through the 3 kHz region and thins out in the bass, purely because of how loudness works. That is the same physics behind the “loudness” compensation knob on older amplifiers.

Beyond perception, each dropout or ring you notice maps to a hardware or hearing limit: laptop speakers roll off below roughly 150–200 Hz and cannot make the first six steps audible at all; room modes ring persistently between about 40 and 200 Hz; Bluetooth codecs low-pass around 16–20 kHz, erasing the top steps; and age-related high-frequency loss removes individual steps in the treble long before the middle range is affected. None of these is a diagnosis; they are locations worth re-checking with a steady tone.

Accuracy, limits and troubleshooting

• Constant level is constant dBFS, not constant loudness or SPL. Digital full scale describes the file, not the air. Do not judge “dropouts” by perceived volume alone; check the analyzer trace at that frequency.
• Grid resolution is capped by your device. The top step is 20 kHz, which needs a sample rate above 40 kHz to exist at all (Nyquist: 22,050 Hz ceiling at 44.1 kHz). A device reporting 32 kHz output cannot play the last several steps, no matter what the page shows.
• Narrow notches hide between steps. With 32 steps the sweep samples 1.32 bands at a time; a sharp room null can fall between grid points. Drop to 48 steps (0.85 bands each) for finer probing, accepting that each tone sounds for barely half a second.
• If the sweep sounds like one long rising buzz, your output device is resampling aggressively or an OS “enhancement” (bass boost, dialogue leveler) is running. Disable effects and try wired headphones.
• Stop when your ears tire. Pitch memory and sensitivity degrade through a session; run short passes, up and down, over separate sittings for consistent readings.

Related tools

• Sweep Generator — continuous linear and logarithmic sweeps for classic rolloff checks.
• Tone Generator — hold any flagged frequency steady for real measurement.
• Frequency Check — a stepped hearing-range test to pair with what the sweep reveals.
• Noise Generator — pink and white noise for broadband response checks without pure-tone ringing.

Aliases covered on this page: perceptual sine sweep, perceptual sweep, and hearing band sweep.