Method
How we measured speaker cleaner sounds
We measured 9 speaker cleaner sound files from 10 sites, and our own tone. We checked the speaker cleaner frequency that popular sites claim. Most claim a tone near 165 Hz. One of the six "water" files plays in that range. This page shows the method and the numbers.
Of 6 distinct "water" sound files, only 1 has its dominant frequency inside 150 to 300 Hz. That file is served by 1 site. The other 5 play 530 or 15 Hz, or mixed signals. One of these 5 is the widely copied file in Fig. 1, which opens with 530 and then holds a 15 Hz square wave. See the study table.
Spectrogram of the measured file (our measurement). Log frequency axis.
What frequency should a speaker cleaner use?
Short answer: no test shows one best frequency for a speaker cleaner. Many speaker cleaner pages say that a sound removes water and dust. The evidence covers water droplets only, and none covers dust. For water, a low frequency at high volume is the stated aim. For dust, no frequency works.
For water, the aim is to move the speaker membrane as far as it can go. Then the moving air pushes on the water in the grille. A Bose research director quoted by The Verge says to play "the lowest tone that that speaker can reproduce, at the loudest level that it can play". This is expert opinion, not a measurement.
The popular 165 Hz figure comes from one 2016 forum post about an iPhone 7, which BGR quoted. We found no data that supports it as a best value. The speaker in a phone has its resonance at about 450 to 900 hertz, so the figure is a low frequency, not a resonance.
Apple Watch plays a series of tones to eject water. Apple does not publish the frequency. An Apple patent says that about 10 tones can be used. Pauses between the tones let the water move again.
Physics papers show that vibration can detach droplets. The best frequency depends on the droplet size and on the surface. We found no paper that tested a speaker from a phone. Claims about sound frequencies for dust give no data.
Does a sound tool work the same on an iPhone and on an Android phone? We found no maker page for a phone that describes a sound feature to eject water. Some watches have one. A tone can push droplets out of the grille. It cannot remove water from the charging port, and it does not clean dust.
What this means for your phone
- Water: use short sessions of low-frequency tones, with pauses. Our tool uses pulses in the range listed below. It is a starting range, not a proven best value.
- To remove water first, tap the phone gently with the connector facing down, and leave it in a dry area with airflow. Apple gives this step for an iPhone. A tone is an extra step, and it does not reach trapped water in the charging port.
- Dust and lint: no frequency removes them. Apple names a small, dry, soft-bristled brush. Dubawa rated a viral video that claimed a sound can clean a speaker as "Misleading".
- Our reading, not a measurement: higher frequencies move the diaphragm very little. We found no evidence that they remove water or dust.
- To clean dust, use a dry brush and a microfiber cloth. General advice: if the sound stays muffled after you clean and dry the phone, ask a professional repair shop.
- Water or lint can muffle the sound. Use the speaker check to hear the playback after you clean it.
- Hold the phone away from your ear. Stop if you hear ringing or feel pain.
Read how to clean dust from a speaker.
We measured 9 speaker cleaner sounds
Of 6 distinct "water" files, 1 has its dominant frequency inside 150 to 300 Hz (the range these sites claim). That file is served by 1 site.
We took 14 audio files from sites on 2026-10-09. We compared checksums (MD5). 10 distinct files remain: 4 of the files are identical, and 2 more are identical. 10 sites serve these files. The widely copied file is served by 5 sites (4 host it and 1 links to a copy). A second file is served by 2 sites.
We excluded 1 file (served by Site J): it is a music or video rip, not a tone file. That leaves 9 distinct tone files: 6 "water" files and 3 files that the sites call "vibrate mode". We label the vibrate files as such. We do not count them in the headline.
We list the sites as Site A to Site K. We do not name them.
| File (Site) | Served by (sites) | Duration | Dominant frequency over time (Hz) | Energy in the 150 to 300 Hz band (%) | Clipping | Notes |
|---|---|---|---|---|---|---|
| Water file 1 (Site A, B, C, D, K) | 5 (4 host it, 1 links) | 74.9 s | 530 (0 to 10 s), 15 (10 to 20 s and 60 to 70 s), a rise from 1200 to 1900 (20 to 60 s), 530 (70 to 75 s) | 0.5 % | 6.1 % | Stereo, channels equal. The widely copied file. |
| Water file 2 (Site E, F) | 2 | 59.1 s | 530 in every window | 0.0 % | 31.7 % | Stereo, channels equal. One constant tone. |
| Water file 3 (Site G) | 1 | 86.7 s | 15 in every window (harmonics at 45, 75, 105) | 2.0 % | 0.0 % | Stereo, channels equal. RMS 0.93. |
| Water file 4 (Site H) | 1 | 3.0 s | 530 | 0.0 % | 26.4 % | Mono. A short loop. |
| Water file 5 (Site A, deep mode) | 1 | 67.3 s | Mostly 95 to 130. Two 5-s rises to about 195 (10 to 15 s and 45 to 50 s). Bursts at 2000 to 6000 (20 to 25 s and 50 to 60 s) | 12.6 % | 6.1 % | Mono. Up to 85.5 % of the energy is above 4 kHz in one window. |
| Water file 6 (Site I) | 1 | 50.0 s | A slow rise from 162 to 198 | 100 % | 0.0 % | Mono. The only water file in the 150 to 300 range. |
| Vibrate file 1 (Site A) | 1 | 60.6 s | 194 to 196 | 95.4 % | 0.0 % | Vibrate mode. Stereo, channels equal. |
| Vibrate file 2 (Site A, deep mode) | 1 | 100.0 s | 7300 (0 to 20 s), then 180 to 215, then 215 (75 to 100 s) | 77.3 % | 0.0 % | Vibrate mode. Mono. 52.6 % of the energy is above 4 kHz in the first 20 s. |
| Vibrate file 3 (Site J) | 1 | 32.1 s | 194 to 196 | 84.4 % | 0.0 % | Vibrate mode. Sample rate 24 kHz, stereo. |
| Our tone, one cycle (reference) | ours | 22.5 s | Pulses at 150 to 300. Dominant: 167, 200, 233, 267, 300 per 5-s window | 93.0 % | 0.0 % | Mono. Rendered from protocol.json. Not a phone recording. |
Energy in the band is the share of spectral energy between 150 and 300 Hz, averaged over the 5-second windows. For the file with 4 sites on the same checksum we measured one copy. All numbers come from our script media/measure/study.py.
Speaker cleaner frequency: what each file plays
Each image shows one distinct file. The frequency axis is logarithmic. The two white lines mark 150 and 300 Hz. We publish the images only, not the audio.
Method update
- Before: we analysed one file. Now we analyse every distinct file that we found. We compare MD5 checksums first, so identical files count once.
- We use the same method for each file: decode at the native sample rate (no resampling), split the channels, cut 5-second windows, apply a Hann window and an FFT, and list the four strongest peaks per window.
- New: for each file we compute the dominant frequency per window, the share of energy in the 150 to 300 band, the share above 4 kHz and above 8 kHz, the clipping fraction, the left and right RMS difference, and the duration.
- "Dominant frequency inside 150 to 300 Hz" means that, in more than half of the windows, the strongest peak is between 150 and 300 Hz.
- We render our own tone from src/data/protocol.json with the same analysis, as a reference row.
- These are the files the sites served on 2026-10-09. Files may change.
- We measure the signal in the file. We do not measure what a phone speaker outputs.
- We anonymised the sites on purpose.
Method
- Sample rate: 44.1 kHz. The third-party file is 44.1 kHz stereo, so we did not resample it for the table.
- Table of findings: we split the left and right channels with ffmpeg. We cut each channel into 5-second windows. We applied a Hann window and an FFT to each window. We listed the four strongest spectral peaks per window. We also counted the share of energy above 4 kHz and above 8 kHz, and the share of samples at clip level (absolute value above 0.99).
- Spectrograms: we mixed the channels to mono. For the third-party file we used a Hann window of 32768 samples (0.74 s) with a hop of 8192 samples. For our tone we used a Hann window of 8192 samples (0.19 s) with a hop of 1024 samples. The colour scale covers 70 dB below the loudest point.
- Tools: ffmpeg, Python, NumPy, and Pillow. The script is in our project as media/measure/make_measure.py.
- We publish an image of our measurement only. We do not publish the third-party audio.
What a widely copied speaker cleaner file plays
The file is a 74.9 s MP3 of 1.94 MB. The page that plays it uses a plain HTML audio player. It does not synthesise a tone in the browser.
Findings
| Item | What we measured |
|---|---|
| Left and right channels | The same. The RMS difference is below 0.001. The file does not target one channel. |
| 0 to 10 s | A 530 Hz tone with harmonics. The strongest peaks are 530, 1590, 1060, and 2650. |
| 10 to 20 s, and 60 to 70 s | A 15 Hz signal with odd harmonics at 45, 75, and 105. This is consistent with a square wave. RMS is 0.96. |
| 20 to 60 s | A slow upward sweep from about 1.2 kHz to about 1.9 kHz, at a lower level (RMS 0.15 to 0.32). |
| 70 to 75 s | 530 Hz again. |
| Clipping | 6.1 % of samples are at the clip level. The file is hard-clipped. |
| High frequencies | Less than 7 % of the energy is above 4 kHz in every window. Less than 1.5 % is above 8 kHz. There is no high-frequency band. |
| 165 Hz | Not among the four strongest peaks in any window of the main file. |
| Second file ("vibrate" mode) | 60.7 s. A tone at 194 to 196 with tremolo sidebands between 150 and 205. RMS is 0.25. No clipping. |
Reconciliation note
An earlier quick check of the same file used a 1-second window. It reported about 34 Hz at 10 s. The 5-second Hann-windowed FFT at the native 44.1 kHz is the governing measurement. It shows a 15 Hz fundamental and the odd harmonics at 45, 75, and 105 as the four strongest peaks. A 1-second window holds only 15 cycles of a 15 Hz signal. Its frequency resolution is about 1 to 2 Hz and it shows strong spectral leakage. A single-bin pick can land on a leakage lobe or an intermodulation bin. We keep the earlier note unedited, and we use the 5-second result.
Why a phone speaker cannot play 15 Hz
Interpretation, not a measurement:
A typical smartphone micro-speaker has a resonance of about 450 to 900 Hz. Industry coverage says its response extends down to below about 800 Hz, and some variants go a little lower. A 15 Hz fundamental is far below this range.
We expect that the speaker cannot reproduce a 15 Hz fundamental. It will probably render the sharp edges of the square wave as a click or buzz train. We have not recorded this on a phone yet.
How common is this file?
In our competitor audit of 15 speaker cleaner sites, seven shared an identical audio file or identical tool code. Five sites used or linked this one file (same checksum). The site text said the tone is "about 165 Hz". The file does not match that text. We do not name the sites. We call this "a widely copied speaker cleaner site template".
Re-creation of what we measured
This is a re-creation, not the original file. We made it ourselves to show what the measured signal sounds like. It has a 530 Hz tone with harmonics, then a 15 Hz square wave, then a sweep from 1.2 to 1.9 kHz. It is about 20 seconds long. Turn the volume down before you press play. The 15 Hz part may sound like a buzz or clicks.
What we play instead
Our tone is generated in your browser from one settings file, src/data/protocol.json. The tool also shows a planned frequency and a measured frequency while it plays.
Spectrogram of our tone, one cycle, synthesised from protocol.json (our measurement of the signal, not a phone recording).
We did not yet record our tone on real phones. The spectrogram of our tone shows the signal we send to the speaker. It does not show what a phone speaker outputs. The device test is pending. We will add recordings and publish the date here.
Values from protocol.json
- Waveform: sine. Cycles per session: 3.
- Pulses per cycle: 10. Each pulse is 1.5 s on and 0.75 s off. One cycle is 22.5 s.
- Pulse frequencies in Hz, stepped in equal steps: 150, 167, 183, 200, 217, 233, 250, 267, 283, 300.
- Fade in and fade out: 10 ms for each pulse. Pause between cycles: 12 s.
- Optional resonance pass: a sweep from 450 to 900 Hz, 4 s, 2 times. The optional resonance pass is off by default and is labelled experimental.
The excerpt is at about half of full scale, so it is not at the loudest level.
Sources
- Knowles 11x15x3.5 mm smartphone speaker datasheet (resonance 800 Hz in 1 cm3, 450 Hz in a baffle; third-party mirror)
- audioXpress, 2015: Characterizing the Microspeaker Industry
- The Verge, Aug 2024: Can a YouTube video really fix your wet phone? (iFixit dye test)
- Apple 108352: How to use Water Lock and eject water from your Apple Watch
- BGR, Dec 2016: Free app ejects water from your iPhone 7 speaker (origin of 165 Hz)
- Apple patent US10595107: Speaker module architecture
- Wang et al. 2019, J. Fluid Mech. 876: Water drops bouncing off vertically vibrating textured surfaces
- Apple 101968: iPhone troubleshooting sound issues (soft brush for debris)
- Dubawa, 2025-07-30: Misleading viral TikTok, sound cannot clean your phone speaker
- Apple 102643: If you see a liquid-detection alert on your iPhone
All other numbers on this page come from our own analysis on 2026-10-09 (audio analysis and competitor audit).