Why does a cut click, and how long should the fade be?
Short answer: the click is the waveform jumping from silence to whatever value the music has at your cut point, and back to silence at the end. That value is rarely zero: across 10,000 random cut points in our test recording the jump was -27.4 dBFS at the median, in a clip whose average level was -22.3 dBFS. A fade turns the jump into a ramp. How long it needs to be:
- 1 ms of fade lowered the click from 2,000 random cuts by 23.6 dB at the median, but 1.8% of cuts still measured above -40 dBFS.
- 2 ms put every one of the 2,000 cuts at or below -40 dBFS. 10 ms put the worst at -54 dBFS.
- 0.1 s, the shortest setting on the audio cutter, put the worst at -74 dBFS. In the file the cutter exported, the click fell from -15.5 dBFS with no fade to -78.7 dBFS.
So: open the audio cutter, set both Fade in and Fade out to 0.1 s, and the click is gone in WAV and in MP3. Use longer fades when the music itself needs an ending, not for the click.
Where the click comes from
Digital audio is a list of sample values, 44,100 per second for each of the left and right sides in the files this site writes. When you cut, the first sample of the new file is whatever value the music had at that instant. Before it there is nothing, which a player treats as zero. If that first sample is 0.386, as it was at the start point in our test cut, the signal goes from 0 to 0.386 in one step of 1/44,100 s. A step contains every frequency at once, so it produces a short, wide-band burst. That burst is the click. The same happens in reverse at the end.
How often does a cut land on a non-zero value? We took 10,000 random cut points in a 30 s orchestral excerpt (details below) and looked at the sample there:
| Measure | Value |
|---|---|
| Level of the sample at the cut, median | -27.4 dBFS |
| Quietest 10% of cut points | below -39.0 dBFS |
| Loudest 10% of cut points | above -16.9 dBFS |
| Cut points where both left and right were below -40 dBFS | 8.3% |
| Cut points where both left and right were exactly zero | 0 of 10,000 |
| Distance to the nearest sample with both left and right below -40 dBFS | 0.48 ms median, 3.74 ms for 90% of points, 50.2 ms at worst |
Cutting exactly at a zero crossing is not a practical way out. You type times to the millisecond, and one millisecond is 44 samples, so you cannot aim at one sample. A stereo file needs both its left and right sides to be near zero at the same sample, which happened at 8% of points in this recording. And the cutter does not move your cut point for you. A fade works at any cut point, so that is what the tool offers.
How we tested
- Music: 30 seconds (1:00 to 1:30) of Mozart's Symphony No. 40, first movement, played by the Musopen Symphony. It is in the public domain on Wikimedia Commons as a 48 kHz, 24-bit FLAC. We converted it to 44.1 kHz, 16-bit stereo.
- Test tone: a 220 Hz sine at -6 dBFS, cut exactly at a crest of the wave (the worst case, where the first sample is 0.5) and, separately, exactly at a zero crossing.
- Fade: the same linear ramp the cutter's code applies, sample by sample, so a 10 ms fade multiplies the first 441 samples by 0/441, 1/441 and so on. Fades of 1, 2, 5, 10, 20 and 50 ms were computed with that formula, because the page's sliders start at 0.1 s.
- Click score: we add 50 ms of digital silence before and after the clip, which is what a player outputs around it, pass everything through a 2 kHz high-pass filter (4th-order Butterworth, zero phase, SciPy), and read the peak of the filtered signal inside the added silence next to the cut. A smooth signal leaves almost nothing there; the uncut tone measured below -150 dBFS. A step rings into it. The music inside the clip is never in the window, so its own treble does not count.
- Real exports: the audio cutter on this site, driven by Playwright 1.63.0 in Chromium 153 on a local copy of the page. We typed the start and end, set the fade sliders, and saved WAV and MP3 downloads. MP3 files were decoded with FFmpeg 4.4.2.
A tone cut at its crest: click level against fade length
The tone has no energy above 220 Hz, so anything the filter finds is the cut.
| Fade in | Click, dBFS | Below the tone by |
|---|---|---|
| none (hard cut) | -12.9 | 6.9 dB |
| 1 ms | -42.0 | 36.0 dB |
| 2 ms | -48.0 | 42.0 dB |
| 5 ms | -56.0 | 50.0 dB |
| 10 ms | -62.0 | 56.0 dB |
| 20 ms | -68.0 | 62.0 dB |
| 50 ms | -76.0 | 70.0 dB |
| 100 ms (cutter minimum) | -82.0 | 76.0 dB |
| 500 ms | -96.0 | 90.0 dB |
| 1 s | -102.0 | 96.0 dB |
The hard cut produces a burst only 6.9 dB quieter than the tone itself. One millisecond of fade removes 29 dB of it, and every doubling of the fade after that removes another 6 dB. Cutting the same tone exactly at a zero crossing, with no fade, scored -39.3 dBFS: better than the crest, but the wave still bends sharply there, and it is about what a 1 ms fade gives. With the 0.1 s fade on top, the zero-crossing cut scored -98.7 dBFS.
2,000 random cuts in music
Real cut points are not all worst cases. We cut the Mozart excerpt at 2,000 random points, applied each fade length to the end of the cut, and scored the click.
| Fade out | Median click | 90% of cuts at or below | Worst cut | Cuts above -40 dBFS |
|---|---|---|---|---|
| none (hard cut) | -33.5 dBFS | -23.1 dBFS | -13.2 dBFS | 75.4% |
| 1 ms | -57.1 dBFS | -45.8 dBFS | -34.0 dBFS | 1.8% |
| 2 ms | -63.2 dBFS | -51.9 dBFS | -40.0 dBFS | 0% |
| 5 ms | -71.1 dBFS | -59.8 dBFS | -48.0 dBFS | 0% |
| 10 ms | -77.2 dBFS | -65.9 dBFS | -54.0 dBFS | 0% |
| 20 ms | -83.2 dBFS | -71.9 dBFS | -60.1 dBFS | 0% |
| 50 ms | -91.1 dBFS | -79.8 dBFS | -68.0 dBFS | 0% |
| 100 ms (cutter minimum) | -97.2 dBFS | -85.9 dBFS | -74.0 dBFS | 0% |
| 500 ms | -111.1 dBFS | -99.8 dBFS | -88.0 dBFS | 0% |
Three out of four hard cuts scored above -40 dBFS, and the worst was -13.2 dBFS, close to the music's own loudest treble. Two milliseconds of fade brought every cut to -40 dBFS or lower. The cutter's shortest fade, 0.1 s, is 50 times that and left the worst cut at -74 dBFS.
What the cutter actually exported
To check that the formula above is what the page does, we cut 12.610 s to 17.637 s of the excerpt on the site, a 5.027 s region chosen because both ends land in loud passages: the first sample was 0.386 on the left (-8.3 dBFS) and the last was 0.320 in the right (-9.9 dBFS).
| Export | Fade in / out | Click at start | Click at end | File size |
|---|---|---|---|---|
| WAV | none | -15.5 dBFS | -17.1 dBFS | 886,808 bytes |
| WAV | 0.1 s | -78.7 dBFS | -85.5 dBFS | 886,808 bytes |
| WAV | 0.5 s | -95.4 dBFS | -105.2 dBFS | 886,808 bytes |
| MP3 128 kbps | none | -16.1 dBFS | -17.8 dBFS | 81,084 bytes |
| MP3 128 kbps | 0.1 s | -64.7 dBFS | -80.5 dBFS | 81,084 bytes |
The WAV files matched our offline formula to within 1 step of 16-bit rounding at every sample, and the hard-cut scores were identical to the offline ones (-15.5 and -17.1 dBFS), so the tables above describe what the page does. A fade changes no file size, because the same number of samples is written.
MP3 adds a twist. Every MP3 this encoder writes begins with 1,105 samples (25.1 ms) of encoder delay before the music. With a hard cut, that stretch was not silent: the encoder smeared the click backwards, and the decoded file rose above -60 dBFS 19.1 ms before the music started, peaking at -21.5 dBFS. After the music, the leftover samples peaked at -28.1 dBFS. With the 0.1 s fades, the lead-in stayed below -55.2 dBFS and rose above -60 dBFS only 3.7 ms before the music, and the tail peaked at -65.2 dBFS. A hard cut in an MP3 is therefore a click plus a short smear on both sides of it.
What a 0.1 s fade costs, and when to go longer
A 0.1 s linear fade multiplies the first 4,410 samples by a ramp from 0 to 1. Over those 100 ms the music is on average 6.0 dB quieter than the original; 10 ms in it is 20 dB down, 50 ms in it is 6 dB down. If you cut so that a note starts exactly at your start point, its attack is softened. Set the start point a little before the note, or in the gap before it, and the fade covers silence or a quiet tail instead.
Longer fades are about music, not clicks. If the cut ends mid-phrase, a 0.5 to 2 s fade out makes an ending; for a clip that repeats, such as a ringtone, our Android ringtone test measured 1 and 2 s fade outs. If Fade in and Fade out together are longer than the region, the page shortens both in proportion so they fit.
- Open the audio cutter and set the start and end of the part you want.
- Set Fade in and Fade out to 0.1 s. Press Play region to hear it.
- Pick WAV if you will edit the clip again, or MP3 for a smaller file, and press Download. Fades and format apply to every region you add.
Questions
Is the click a fault in the file or in the cutter?
Neither. It is the step between silence and the first sample, and every cutter that keeps your exact cut point produces it. In our test the sample at a random cut point was below -40 dBFS on both left and right at only 8.3% of points, so most hard cuts will have one.
Does exporting MP3 hide the click?
No. The hard-cut MP3 scored -16.1 dBFS at the start against -15.5 dBFS for the WAV of the same cut, and the encoder spread part of the click into the 25.1 ms lead-in, which peaked at -21.5 dBFS. With 0.1 s fades the same lead-in stayed below -55.2 dBFS.
Is 1 ms enough?
Not for every cut. At 1 ms, 1.8% of the 2,000 random cuts still scored above -40 dBFS and the worst was -34.0 dBFS. At 2 ms none did. The cutter's sliders start at 0.1 s, which left the worst cut at -74 dBFS.