Sine Sweep Generator: 20 Hz to 20 kHz, Log or Linear

A sine sweep is one pure tone that glides from a low frequency to a high one, so every pitch between 20 Hz and 20 kHz plays in turn. By ear, it finds the note that makes a panel rattle, a room boom or a speaker drop out. The generator below plays a log or linear sweep of 1 to 120 seconds on your device, shows the frequency leaving the speakers as it plays, and saves the same sweep as a WAV.

How to run a sine sweep

1. Turn the volume down. Set the system volume low before the first press. The tool starts at -24 dBFS peak and stops at -12 dBFS. That's a digital level, not a reading in dB SPL, so the volume knob on your amp, laptop or phone still decides how loud the room gets.
2. Pick a range. Tap one of the five quick ranges (full range, bass and sub, slow bass up and down, mids, top end) or type Start and End in hertz. Put Start above End for a falling sweep.
3. Set law, length and pattern. Log gives every octave the same time and suits listening; Linear gives every hertz the same time. Length runs from 1 to 120 seconds per pass. Once plays a single pass, Loop repeats it, and Up and down sweeps back to where it started.
4. Press Start. The readout shows the frequency, the time, the direction and the pass. Level and channel change while the sweep plays; range, law, length, pattern and fade wait for the next Start. Pause freezes the sweep, Resume carries on from the same frequency, and Stop or the Escape key ends it.
5. Mark what you hear. Press Mark this frequency, or M on the keyboard, the moment a rattle or boom starts. Each mark keeps the frequency, the time into the pass and whether the sweep was rising or falling, and the last 12 stay on the list. Type a frequency into the finder to see when it plays.
6. Save the WAV. Optional: choose mono or stereo, 44.1 or 48 kHz and 16-bit or 32-bit float, then press Save WAV. The file holds the same sweep at the current level, fade and channel, and the size line shows its bytes before you save.

Range, law, length, pattern, channel, fade and the WAV choices are remembered on this device. The level isn't. Every visit starts back at -24 dBFS peak, which is the safer place to begin when you don't remember how loud the amp was left.

What a sweep tells you by ear

A sweep is the quickest way to find a frequency that misbehaves. Only one tone plays at a time, so a resonance stands out as a jump in loudness, a buzz or a rattle at one spot and nowhere else. That's why audiocheck's page of 20 Hz to 20 kHz test sweeps recommends sweeps for finding resonant frequencies.

Expect the loudness to change even on a good system. Hearing isn't flat: the equal-loudness contours show the ear is most sensitive between 2 and 5 kHz, so a speaker with a flat response still seems to swell through the upper mids and thin out toward both ends. A broad, smooth swell is your hearing. A sharp jump that comes and goes within a few hertz is the thing to mark, and the audio frequency chart shows which instruments and voices live where the mark landed.

Hearing a sweep isn't measuring it. A frequency response graph needs a measurement mic and software that compares what the mic recorded with the exact signal it sent. REW's getting-started help describes exactly that setup: it measures with its own log sweep, and for full-range results the mic response must be calibrated. To see roughly where peaks sit without REW, record the sweep in the room with your phone and open the recording in the audio spectrum analyzer. A phone mic isn't calibrated, so trust where a peak sits more than how tall it looks.

Log or linear sweep: which to pick

Log is the one to listen to. A log sweep takes the same time to double in frequency, so every octave gets an equal share of the run. A linear sweep moves the same number of hertz every second, which makes the bass fly past: on a linear scale, audiocheck notes, 20 to 40 Hz takes as long as 10 to 10.02 kHz. The table shows what that means in a 20-second full-range sweep.

Half of a linear run sits in the top octave, where the tweeter does the work and the ear's sensitivity is already falling away. The whole bass range below 200 Hz gets less than a fifth of a second. Played to the ear, a log sweep moves at an even pace from bottom to top, and audiocheck describes its spectrum as pink-like and a linear sweep's as white-like. Linear has one real job: feeding an FFT analyzer whose frequency bins are evenly spaced in hertz.

Length trades time for detail. REW's measurement help says each doubling of sweep length improves signal-to-noise by almost 3 dB, which matters for a measurement. By ear, a longer sweep simply gives you more time to catch the note. The bass preset spends 9.0 s on each octave between 20 and 200 Hz, and the slow preset spends 20.6 s per octave on 20 to 150 Hz and then comes back down, so a rattle gets two chances to show itself.

My pick is a 20-second log sweep for the first full-range pass, then a slow bass sweep over whatever misbehaved. I'd switch to linear only when an FFT analyzer is reading the sweep, because that's the one case where even spacing in hertz helps.

Sweep recipes for common checks

Find a rattle or buzz

Run the Bass and sub preset at a moderate level and walk the room, or put your ear near the panel, shelf or grille you suspect. Press M when the buzz starts and again when it stops. Then hold the note: set the online tone generator to the marked frequency, and the rattle keeps going while you press a hand on each surface until it stops. The bass test does the same job with fixed steps, which is easier when you'd rather tap through tones than chase a moving one.

Hear room resonances

Rooms boom at predictable notes. The strongest come from axial modes between two parallel walls, which the room modes article describes as the ones with the most influence on a room's low-frequency response. The first sits at the speed of sound divided by twice the distance. With sound at 343 m/s, its speed in air at about 20 °C, a 5 m room length gives 34.3 Hz, then 68.6 and 102.9 Hz. A 4 m distance gives 42.9 Hz, and 6 m gives 28.6 Hz.

I'd run the slow bass preset from the listening seat, since that's where the boom matters. Mark each note that swells and falls away, then compare the marks with your room's dimensions. A mode is louder in some spots than others, and EQ fixes it for one seat only, so moving the seat or the speakers is the cheaper first try. The speaker placement calculator gives a starting layout for a stereo pair.

Check a range by ear

For headphones and small speakers, the Top end preset (5 to 20 kHz over 15 s) shows where the treble fades out for you. That point mixes three things: the driver, the device and your ears. The human hearing range is commonly given as 20 Hz to 20 kHz, with wide variation between people at the top and a gradual loss with age, so a sweep that goes quiet before its end is normal. Switch the channel to Left or Right to hear one side at a time; the headphone test and the left and right speaker test are quicker when you only need to know that both sides play.

Record a sweep through gear

The saved WAV is a clean source for checking a cable, an interface or a plug-in chain. Play it through, record the output, and listen or look for anything that wasn't in the file. 16-bit is smaller and opens anywhere; 32-bit float keeps more precision for a chain you'll process further. A 20-second full-range sweep at 48 kHz stereo is 3,840,044 bytes as 16-bit and 7,680,058 bytes as 32-bit float, and the audio file size calculator works out other lengths and formats.

One place this WAV won't work is REW's file playback mode. REW's help says that mode needs a measurement sweep generated and saved from REW's own signal generator, with its acoustic timing reference, so make the file there.

Keep tweeters and ears safe

A sweep is harder on speakers than music or noise at the same peak level. The reason is energy: as audiocheck's notes on sweeps put it, a sine sweep gives the full available level to one frequency at a time, where noise spreads it across many. Tweeters take that badly. In his article on why tweeters blow when amplifiers distort, Rod Elliott of Elliott Sound Products points out that there's very little airflow around a tweeter's voice coil, so the heat has nowhere to go, and that an amplifier pushed just 3 dB into clipping can feed a tweeter up to 3 times the power it saw before clipping. Sustained, that overheats the coil and damages its adhesive, its former and the enamel insulation on the wire.

My rule: start at the default level with the system volume low, raise the volume only until the middle of the sweep is comfortable, and back off the moment anything distorts. Keep long sweeps quieter than short ones. REW's own help warns that the risk of overheating a tweeter grows with sweep length.

The cap sits 3 dB under REW's default on purpose. This page is for listening. A measurement run belongs in REW with a calibrated mic, and REW's help notes that its maximum level puts the sweep's peaks at digital full scale.

Your ears need the same care. The NIDCD's page on noise-induced hearing loss says sounds at or below 70 dBA are unlikely to cause hearing loss even after long exposure, while long or repeated exposure at or above 85 dBA can cause it, and the louder the sound, the shorter the time it takes. The decibel meter gives a rough dBA reading from a phone or laptop mic at your seat. Through headphones, keep the sweep at a level you could talk over.

For matching the level of two speakers, pink noise is the gentler signal, since it spreads its energy across the band instead of parking it on one note. The pink noise generator is built for that job.

Limits from your device and browser

The top of a sweep is capped by the sample rate. A digital system can't carry anything above half its rate, the Nyquist frequency: 22,050 Hz at 44.1 kHz and 24,000 Hz at 48 kHz. The live sweep runs at the sample rate of the browser's audio context, which the readout shows after Start. The tool stops a little short of Nyquist, at 49 percent of the rate: 21,609 Hz at 44.1 kHz and 23,520 Hz at 48 kHz. A higher End frequency is clamped to that limit, and the status line says so.

Bluetooth adds a codec in the path. SBC is the default codec every A2DP device must support, and it was designed around Bluetooth's bandwidth limits to give reasonable quality at medium bit rates, so the sweep you hear has passed through SBC, or whichever codec the two devices agree on, on its way to the speaker. The readout follows the output position the browser reports, and the output latency behind it is an estimate. On wireless output, read the frequency as approximate and switch to a wired connection when the exact note matters. In a 20-second full-range log sweep, every 0.1 s of unreported delay puts the readout about 3.5 percent, a little over half a semitone, ahead of the note you hear.

Windows audio enhancements process the signal before it reaches the speakers, which can skew a sweep. Microsoft's guide to fixing sound problems says enhancements can cause audio issues and shows how to switch them off. In Windows 11, open Start > Settings > Sound, select the output device, scroll to Advanced settings and set Audio enhancements to Off. In Windows 10, open Control Panel > Hardware and Sound > Sound, open the Properties of the default playback device, and on the Enhancements tab tick Disable all enhancements (or Disable all sound effects).

Nothing plays until you press Start. Browsers hold back audio that starts outside a user action, as MDN's autoplay guide explains, so the tool opens its audio output only from the button.

The largest WAV the tool can build, a 120-second up-and-down sweep in 32-bit float stereo at 48 kHz, is 92,160,058 bytes, and the browser holds all of it in memory while building it. When a phone runs out, the status line suggests mono, 16-bit, 44.1 kHz or a shorter sweep, and any one of those cuts the size.

How the generator builds the sweep

The live sweep is one sine oscillator whose frequency the browser ramps on its audio clock. Log mode uses exponentialRampToValueAtTime, which MDN documents as an exponential change between two positive values, the same curve a log sweep follows. Linear mode uses linearRampToValueAtTime. An exponential ramp can't reach zero, so the tool needs a floor above it and sets that floor at 10 Hz.

Pause fades out over 30 ms and then suspends the audio context, which halts the clock the sweep is scheduled on, so Resume picks up within a few hundredths of a second of where the sound faded instead of jumping ahead. The readout uses getOutputTimestamp, the position the output device is playing, and falls back to the context time minus the reported output latency in browsers without it.

The WAV isn't a recording of the live sound. It's computed sample by sample from the closed-form phase of the sweep given in Angelo Farina's AES paper on exponential sweeps, so the file has no steps or clicks however long it runs. The same paper shows why measurement software prefers the exponential sweep: after recording, deconvolution separates the linear response from each order of harmonic distortion. That needs the recording and the exact source sweep together, which is what REW does with its own file. 16-bit files are plain PCM with a 44-byte header. 32-bit float files use format code 3 with an 18-byte format chunk and a fact chunk, a 58-byte header, as the WAVE format notes from McGill describe. The browser builds the file in chunks, saves it to your device and releases its object URL a minute later.

My verdict: use this page for the by-ear jobs, finding a rattle, a room boom or the top end of a pair of headphones, with a log sweep at a low level. For a frequency response graph, use REW's own sweep with a calibrated measurement mic: a graph needs software that knows the exact sweep it sent and a mic whose errors are known, and listening gives you neither.

Sine sweep generator FAQ

Why can't I hear the top of the sweep?

Usually one link in the chain stops before 20 kHz: the speaker or earbud, the device, or your ears. Human hearing is commonly given as 20 Hz to 20 kHz, with wide variation at the top and a gradual loss with age, so a sweep often goes quiet before it ends. The tool itself stops at 49 percent of the sample rate, 21,609 Hz at 44.1 kHz.

Can a sine sweep damage speakers?

Yes, at high level or through a clipping amplifier, and tweeters are the weak point. A sweep puts all its level into one frequency at a time, a tweeter's voice coil sheds heat poorly, and an amp pushed 3 dB into clipping can feed it up to 3 times the power it handled before clipping. Keep the system volume low, start at the default -24 dBFS peak and stop when you hear distortion.

Can I measure my room's frequency response with a sweep?

Not by ear. A response measurement needs a calibrated measurement mic and software such as REW that compares the recording with the exact sweep it sent, using a sweep from its own generator. By ear, a slow bass sweep from the listening seat still finds the notes where the room booms.

Should I use a log or a linear sweep?

Log for listening. It spends the same time on every octave, 2.01 s each in a 20-second full-range sweep. Linear spends half of that sweep above 10 kHz and suits FFT analyzers whose frequency bins are evenly spaced in hertz.

Why do some parts of the sweep sound louder?

The ear's sensitivity changes with frequency, and it peaks between 2 and 5 kHz, so even a flat speaker seems to swell in the upper mids and thin out at both ends. A sharp jump at one note points to a resonance; a broad, smooth swell is mostly hearing.

Can I use the WAV in a DAW or in REW?

In a DAW, yes: it's a standard 16-bit PCM or 32-bit float WAV at 44.1 or 48 kHz. REW's file playback mode needs a measurement sweep generated by REW itself, with its timing reference, so use REW's generator for that.

Does the tool record or upload anything?

No. The sweep is generated on your device with the Web Audio API, the WAV is built in the browser, and the tool never asks for the microphone. Only your settings are remembered on this device, and the level resets on every visit.