How long should a fade out be, and why does it seem to drop off at the end?
Short answer: there is no one right length, but there is a reason a fade out can seem to hang on and then cut off. The audio cutter's fade is linear: the volume falls in equal steps of amplitude, not of loudness. We exported fades of 3, 5, 8 and 10 s and measured them:
- Halfway through, the music is only 6 dB down (-6.03 dB measured, on every length).
- At 90% of the fade it is 20 dB down, so the last tenth goes from -20 dB to silence. On a 10 s fade that last stretch is 0.99 s; on a 3 s fade, 0.29 s.
- The audible drop, from -6 dB to -20 dB, took 1.21 s in a 3 s fade, 2.00 s in 5 s, 3.21 s in 8 s and 4.01 s in 10 s.
- Fading twice (save the faded WAV, open it again and apply the same fade) reached -6 dB after 2.92 s instead of 5.00 s on a 10 s fade, so the drop starts earlier and is spread out more evenly.
Settings: in the audio cutter, put the end marker where the music should be silent and drag Fade out (0 to 10 s, in 0.1 s steps). If it still sounds too sudden, use a longer fade, up to 10 s, or fade twice through a WAV. The fade in and fade out together must fit inside the region, or both get shortened.
Why a linear fade sounds late
A fade out multiplies each sample by a gain that goes from 1 to 0. In the cutter the gain falls on a straight line: a quarter of the way through it is 0.75, halfway 0.5, at 90% 0.1. Level in decibels is 20 × log10(gain), and our hearing judges loudness changes much more like decibel steps than like amplitude steps. On a decibel scale, a straight-line gain is a curve that stays flat for a long time and then plunges: 0.5 is only -6 dB, 0.1 is -20 dB, and 0.01, reached at 99% of the fade, is -40 dB.
So most of the time you set is spent near full volume, and most of the decibel drop is squeezed into the last moments. The fade does reach silence exactly where you put the end marker, but it gets there late.
Applying the fade a second time multiplies the gain by itself (0.5 × 0.5 = 0.25 at the midpoint), which doubles every decibel figure: -12 dB halfway, -40 dB at 90%. A third pass triples them. That is the only way to change the shape with this site's tools, since the cutter offers one fade curve.
How we tested
- Page: a local copy of this site's audio cutter in Chromium 153 driven by Playwright. For each file we typed the start and end, set Fade in to 0 and Fade out to 0, 3, 5, 8 or 10 s, and saved WAV (and once MP3 at 192 kbps). The files are what the browser downloaded.
- Test tone: a 1 kHz sine, 30 s, stereo, 16-bit 44.1 kHz, made with FFmpeg 4.4.2 (-30.1 dBFS RMS). It has a steady level, so any level change is the fade.
- Music: 60 seconds (1:00 to 2:00) of Mozart's Symphony No. 40, first movement, played by the Musopen Symphony, public domain on Wikimedia Commons, as a 16-bit 44.1 kHz stereo WAV. The fade was put on the end of the excerpt.
- Check against the formula: every exported WAV sample was within 1 step (1 LSB) of the source times the straight-line gain; the last sample of every fade was exactly 0. The twice- and three-times-faded files were within 2 and 3 LSB of the gain squared and cubed.
- Level: RMS in 50 ms windows (tone) or 400 ms windows (music), compared with the same stretch of an export with no fade. The MP3 was decoded with FFmpeg 4.4.2 and lined up with the WAV (it starts 1,105 samples late, the encoder delay).
How far down the level is at each point of the fade
| Point in the fade | Gain | Formula | Measured, linear (3 to 10 s) | Measured, faded twice (10 s) | Measured, faded three times (10 s) |
|---|---|---|---|---|---|
| 25% | 0.75 | -2.50 dB | -2.50 dB | -5.01 dB | -7.51 dB |
| 50% | 0.5 | -6.02 dB | -6.03 dB | -12.06 dB | -18.11 dB |
| 75% | 0.25 | -12.04 dB | -12.05 dB | -24.16 dB | -36.45 dB |
| 90% | 0.1 | -20.00 dB | -20.03 to -20.04 dB | -40.42 dB | -63.42 dB |
| 95% | 0.05 | -26.02 dB | -26.06 to -26.09 dB | -53.68 dB | below 16-bit |
The length of the fade changes how many seconds each stage takes, not the shape. Every linear fade is 6 dB down at its midpoint.
Seconds to each level, by fade length
| Fade out | -3 dB | -6 dB | -10 dB | -20 dB | -40 dB | -20 dB to silence |
|---|---|---|---|---|---|---|
| 3 s | 0.89 s | 1.50 s | 2.06 s | 2.71 s | in the last 50 ms | 0.29 s |
| 5 s | 1.47 s | 2.50 s | 3.42 s | 4.50 s | 4.96 s | 0.50 s |
| 8 s | 2.33 s | 4.00 s | 5.47 s | 7.21 s | 7.92 s | 0.79 s |
| 10 s | 2.92 s | 5.00 s | 6.83 s | 9.01 s | 9.90 s | 0.99 s |
| 10 s, faded twice | 1.58 s | 2.92 s | 4.38 s | 6.83 s | 8.98 s | 3.17 s |
| 10 s, faded three times | 1.08 s | 2.06 s | 3.19 s | 5.36 s | 7.82 s | 4.64 s |
Read across the 10 s row: the first 5 seconds take the music down only 6 dB, the next 4 take it to -20 dB, and the last second takes it the rest of the way. Fading twice turns that into roughly 3, 4 and 3 seconds.
The same fades on music, and as MP3
Music does not hold a steady level, so its own loud and quiet passages ride on top of the fade. The unfaded excerpt measured -20.86 dBFS RMS over its last 10 seconds.
| Point in the fade | 10 s fade, against no fade | 10 s fade, actual level | 8 s fade, WAV | 8 s fade, MP3 192 kbps | 8 s, faded twice |
|---|---|---|---|---|---|
| 25% | -2.6 dB | -23.5 dBFS | -2.4 dB | -2.7 dB | -4.9 dB |
| 50% | -6.0 dB | -26.9 dBFS | -5.8 dB | -6.1 dB | -11.6 dB |
| 75% | -11.8 dB | -41.6 dBFS | -12.7 dB | -13.0 dB | -25.4 dB |
| 90% | -19.3 dB | -36.3 dBFS | -18.8 dB | -19.1 dB | -37.6 dB |
| 95% | -26.3 dB | -55.4 dBFS | -26.0 dB | -26.2 dB | -51.5 dB |
- Against no fade the music follows the formula to within 1.2 dB; the wider 400 ms window smooths the curve a little.
- The actual level jumps around: at 90% of the 10 s fade the music was louder (-36.3 dBFS) than at 75% (-41.6 dBFS), because the orchestra got louder there. A fade over a crescendo can seem to stall; a fade over a quiet passage can seem to finish early.
- MP3 at 192 kbps kept the fade: every point was within 0.3 dB of the WAV.
A trap: fades that don't fit the region
The fade in and fade out share the region. We cut a 5 s region and set both sliders to 10 s. The labels next to the sliders still read "10.0 s", but the page scaled both down to fit: the export had a 2.5 s fade in and a 2.5 s fade out (gain 0.5 at 1.25 s, as a 2.5 s linear fade gives). If your fade sounds shorter than the slider says, check that fade in plus fade out is no longer than the region.
How to set a fade out
- Open the song in the audio cutter and put the end marker where it should be silent. The fade ends there and starts the fade length earlier, so for a 10 s fade, the music you want to hear fading needs 10 s before the marker.
- Drag Fade out and press Play region. Our numbers can't tell you what sounds right; we did not run a listening test. They tell you where the drop happens: in the last half of the fade, mostly in the last third.
- If it seems to hold on and then vanish, try a longer fade first. The audible drop from -6 to -20 dB lasts about 40% of the fade (4.01 s of a 10 s fade).
- If 10 s is still not enough, or you want the drop to start sooner, fade twice: set the fade, choose WAV and save; open that WAV, keep the same end and fade length, and save again, as MP3 if you need it. We tested the in-between file as WAV only; WAV stores the faded samples without adding the 1,105-sample delay an MP3 adds at the start.
- Ending a ringtone or a short clip instead of a song? A short fade is enough there: see why a cut clicks (a click needs only milliseconds of fade) and the Android ringtone guide.
Questions
Does the fade out go all the way to silence?
Yes. In every export the last sample was exactly 0, so the file ends in silence at the end marker.
Is the fade in linear too?
Yes, the cutter uses the same straight-line gain for both, so a fade in is also only 6 dB below full at its midpoint and spends its first tenth climbing out of silence to -20 dB.
Does saving as MP3 change the fade?
Not in a way the numbers show: the 192 kbps MP3 was within 0.3 dB of the WAV at every point we measured. The MP3 starts 1,105 samples (25 ms) later, which shifts the whole file, fade included.