Cut a section out of a track on a real waveform, with the two points where you put them and the result on screen before you save. An MP3 is copied out frame by frame instead of being decoded and encoded again, so what you get back is the audio that went in.
Drop an audio file here, or click to choose
MP3, WAV, M4A, AAC, OGG, FLAC, Opus and WebM. A three hour recording opens as fast as a ringtone, because nothing is uploaded.
Free · no sign-up · nothing uploaded. There is no 250 MB cap, because there is no server paying for the bandwidth. An MP3 is cut on its own frame boundaries, so the audio you keep is the bytes that were already in the file.
Reading the file…
MP3, WAV, M4A, AAC, OGG, FLAC or Opus. Nothing is uploaded, so a two hour recording opens without a progress bar.
Pull the handles, or type the times, or press I and O while it plays. Zoom in to put a cut between two words, and loop the selection to check it.
The panel says whether the export is lossless, what it will weigh, and why if it is not. An MP3 comes out as the frames that went in.
An MP3 cutter has two ways to do the job. It can decode the whole file to samples, throw away the part you did not want, and encode what is left. Or it can find the frame boundaries either side of your selection and copy those bytes out.
Almost every one online does the first. That second encode is a second generation of loss, applied to audio you never asked to have re-processed, and it happens even when you are trimming one second of silence off the end of a three minute track. The frequencies the first encoder threw away are gone either way. The second pass then goes to work on what survived, and picks a slightly different set of things to discard.
This one does the second, whenever it can. It is not a better encoder, it is no encoder at all. The badge above the save button tells you which path you are on before you commit to it, and what it cost is measured over ten generations on the guide, on a file you can reproduce.
Land the cut exactly where you put it. An MP3 frame holds 1152 samples and cannot be split, so the in-point moves to the nearest frame boundary. At 44.1kHz that is a quantum of 26.122 milliseconds, and at 48kHz it is 24 milliseconds.
In practice that is under half the length of a spoken consonant, and well inside the gap between two words. It matters for exactly one job, which is cutting a loop that has to line up to a beat, and for that the answer is to save as WAV and accept the size. Everything else is better served by the copy.
The tool never rounds this quietly. The toast after you save says how far the cut actually landed from where you put it, in milliseconds, and it says nothing when the two agree.
| Way | Reach for it when |
|---|---|
| Drag on the waveform | You can see where the section is. Dragging on empty waveform starts a new selection, dragging inside moves the one you have, and a double-click selects the whole file again. |
| Type a time | You already know the numbers, from a transcript or a note. The fields take 1:23.500, 83.5 and 0:83 and all three mean the same thing. |
| Press I and O | You are listening for the point rather than looking for it. Play it, and press the key at the moment it should start or stop. |
| Nudge with the arrows | The point is nearly right. Ten milliseconds a press, a full second with Shift, and one millisecond with Alt held. |
| Snap the length | The far end is fixed by whatever you are making. Thirty seconds for an iPhone ringtone, forty for Android, fifteen for a Reel. |
| Set the length field | You know how long the clip has to be and where it starts. Typing into Length moves the end point and leaves the start alone. |
Every online cutter draws the whole file across one screen width. On a three minute track at 1200 pixels that makes one pixel worth 150 milliseconds, so the finest cut you can place by dragging is six times coarser than the frame quantum you are cutting on. The precision is there in the format and the interface throws it away.
Scroll on the waveform to zoom around the pointer, or use the slider. Past 1× an overview strip appears underneath showing where you are in the whole file and where the selection sits, because zooming without one is how people get lost. Drag on that strip to move the window. Press 0 to fit the file back on screen.
A fade has to be multiplied into the samples, and samples only exist after a decode. So setting a fade above zero turns the lossless path off and the badge says so on the spot, naming the fade as the reason. The file then comes out as a WAV.
Both fades are equal-power rather than linear, which means the curve is a quarter sine instead of a straight line. A linear fade sounds like it dips in the middle, because loudness follows power and power follows the square of amplitude. It is the difference between a fade that sounds smooth and one that sounds like a hole.
If all you want is to stop a clip ending on a hard edge, try moving the out-point into a quiet moment first. That keeps the copy, and on most material it sounds better than a fade anyway.
Not when it does not have to. An MP3 is a queue of self-contained frames, each holding 1152 samples, and cutting one means finding the frame boundaries either side of your selection and copying those bytes straight out. No decoder runs and no encoder runs, so the audio in the file you save is the same bytes that were in the file you opened. The badge above the save button says Lossless when that is what is happening. It changes to Re-encoded the moment you ask for something a copy cannot do, and it says which setting caused it.
Two things force a re-encode: adding a fade, which has to be mixed into the samples, and saving something that is not an MP3 as an MP3, which no browser can do because none of them ship an MP3 encoder. In both cases the file is written as a WAV instead, and the tool says so before you press anything.
On an MP3, to the nearest frame. That is 26.122 milliseconds at 44.1kHz and 24 milliseconds at 48kHz, because a frame is 1152 samples and a frame cannot be cut in half. If your in-point falls inside a frame the whole frame is kept, and the toast after you save tells you how far the cut actually landed from where you put it. On a WAV it is exact to the sample, because a WAV has no frames to land on.
If you need an MP3 cut to a tighter tolerance than a frame there is no way to do it without decoding and re-encoding, and that is a real trade rather than a limitation of this page. Twenty-six milliseconds is under half the length of a spoken consonant, so for trimming silence off a podcast or pulling a chorus out of a song it is not the thing that will bother you.
Not on the cut. There is no upload, so there is no bandwidth anyone is paying for and no 250 MB cap of the kind the site currently ranking first for this imposes. The one real limit is the waveform: drawing it means holding every sample of the file in memory at four bytes each, and past about three hours of stereo audio that is more than a browser tab can carry. A file longer than that is refused with a sentence rather than left to crash the tab.
For everything shorter, a two hour recording opens as fast as a ringtone once the decode is done, and the cut itself is instant because it is a copy rather than a conversion.
MP3 if you opened an MP3 and have not added a fade, and WAV in every other case. That is the honest shape of what a browser can write. Every browser can decode MP3, AAC, FLAC, OGG and Opus, and not one of them can encode any of those, so a tool promising to hand you back an M4A from a WAV is either uploading your file somewhere or shipping a megabyte of encoder to do it.
The WAV path is still lossless with respect to the decode: nothing is thrown away, which is exactly why the file is large. Sixteen-bit is the default and the right choice for almost everything; 24-bit and 32-bit float are there for material that arrived at those depths.
Yes. Switch The selection is to What I cut out and the region turns red and hatches itself, and the two pieces either side are joined into one file. It stays lossless on an MP3 for the same reason a normal trim does, because both surviving runs of frames are copied rather than remade.
The join is a butt splice with nothing smoothed across it, so if the cut lands in the middle of a sustained note you may hear it. Putting the two points in a gap, which the waveform makes easy to find, avoids that entirely. A short fade either side would hide it, and that is the one case where accepting a re-encode is worth it.
Yes, and the handles are sized for a thumb rather than a mouse. The waveform, the two time fields and the transport all work by touch, dragging inside the selection moves it, and pinch-free zoom is on the slider rather than a gesture so it does not fight the page. The keyboard shortcuts are hidden below 620px, because they describe keys the device does not have.
The one thing a phone changes is memory. A tab on a phone has far less of it than a tab on a laptop, so the three hour ceiling is nearer one hour in practice on an older handset. It fails with a message rather than silently.
There is nowhere to keep it. The file is read by your browser with FileReader and decoded by the Web Audio engine already built into it, and every byte stays in the tab. There is no upload, no request and no account, and you can watch that yourself in the network panel: open it, drop a file in, and nothing appears.
The practical version of that is you can turn the wifi off, load the page, and it still works. The first tool ranking for this term stores what you give it in a temporary folder on a server for two hours. This one cannot, because there is no server for it to store anything on.