Audio and MP3 Cutter

Keep the part you need, or delete the middle and join the ends: nothing is uploaded

This audio and MP3 cutter either keeps just the part of a recording or song you pick as a new file (Keep), or removes the part you pick and joins what is left on either side (Delete). Drag on the waveform to select a range, then fine-tune the start and end to 0.01 s in the magnifiers, which enlarge 1.5 s on each side of the edge and show the sound level. Before you save, you can see the result length and the file size down to the byte.

The example on the first screen (a 12-second synth guitar at 44.1 kHz stereo) keeps 6.000 s, from 2.5 s to 8.5 s, and fades out the last second. Saved as a 192 kbps MP3 it comes to 144,822 bytes; saved as a 16-bit WAV it is 1,058,444 bytes. Reading the file, cutting it and building the file to save all happen inside this browser, and the audio itself is never sent to a server.

When do you need to cut audio?

• Ringtones and alert sounds: keep just the chorus of a song. According to Apple, an iPhone ringtone can be up to 30 seconds, so typing 30 in the length box selects exactly 30 s from the start.
• Background music for a presentation or video: trim the track to a set length and add a fade out so it does not stop abruptly.
• Lectures and meeting recordings: keep only the part you want to share. Files up to 30 minutes can be opened, so split longer recordings first.
• Dead air at the start and end: one click on 'Trim silence' removes the quiet right after you pressed record and the noise after you finished.
• A mistake in the middle: a stumbled sentence, a cough or an announcement can be removed with 'Delete selection', and the two sides are joined.

How to use it

1. Choose a task: 'Keep selection' keeps only the part you pick; 'Delete selection' removes it and joins the rest.
2. Open a file: click 'Open file' or drop a file on the workspace. One file is open at a time, and once the waveform is drawn the whole file starts out selected.
3. Select a range: drag across the waveform to select a new range. Drag either orange handle to move only that edge, or drag the middle of the selection to move it without changing its length. A short click on the waveform sets the playhead there.
4. Fine-tune the edges: the two magnifiers below enlarge 1.5 s on each side of the start and the end. Click or drag a magnifier to move the edge there, use the − and + buttons to nudge it 0.01 s at a time, or type a time in the box.
5. Listen and check: 'Play result' plays the result with fades applied, and when deleting, 'Play join' plays 3 s before and after the join. 'Play original' plays the original from the playhead, and if you click 'Use playhead' while listening, that moment becomes the edge.
6. Save: choose a format and quality, click 'Cut to MP3' (or 'Cut to WAV'), then click 'Download'. A player appears in the result area so you can listen to the file you made.

Setting the edges: dB in the magnifier and finding a quiet spot

If you cut while sound is playing, the waveform stops or starts abruptly and tends to leave a short click. So the header of each magnifier shows the loudest level among the 10 ms slice containing the edge and its neighbors, in dBFS (a scale where the digital maximum is 0 dB). The ranges below are the labels this tool uses.

'Snap to quiet' looks within 1.5 s on either side of the edge for the lowest level among the 10 ms slices, and moves the edge to the middle of the slice closest to the current edge among those within 1.5 dB of that lowest level. If nothing is more than 3 dB quieter than the current edge, it does not move. In speech this often lands in a short pause between words, and in music in a weak spot between beats, but always listen after it moves.

'Trim silence' treats the quietest 10% of the 10 ms slices across the whole file as background noise and uses a threshold 10 dB above it (limited to between −60 and −35 dBFS). The start is placed 0.05 s before the first point above the threshold, and the end 0.1 s after the last. In the example file, what remains runs from 0.950 s to 10.880 s.

Keep and Delete: how the result length is decided

Ranges are handled as sample numbers. The time you enter is multiplied by the sample rate and rounded to the nearest whole number; the start sample is included and the end sample is not. So 2.5 s to 8.5 s in a 44.1 kHz file runs from sample 110,250 up to (but not including) sample 374,850, which is 264,600 samples (6.000 s).

Keep: result length = end − start
Delete: result length = original length − (end − start)

When you delete and the start is 0 or the end is the end of the file, there is nothing on one side to join, so the result is the same as keeping the other side. The shortest range is 0.01 s, and if less than 0.01 s of audio would remain after deleting, you cannot save.

Fades and the join crossfade

With Keep, a fade in starts the first sample of the result at 0x and raises it in a straight line to 1x over the time you set, and a fade out lowers it in a straight line so the last sample is 0x. The midpoint of a 1-second fade out (0.5 s before the end) is 0.5x, about −6 dB. If the range is short and the two fades overlap, the two gains are multiplied, so the middle gets quieter too, and if one fade is longer than the length kept, saving is blocked. On the waveform, the faded part is drawn tapered.

With Delete, the join gets a crossfade. Centered on the join, the sound before it is lowered along a cos curve while the sound after it is raised along a sin curve. This equal-power method keeps the level nearly constant in the middle even when two different sounds overlap. With a 0.5 s join, the two sounds overlap for 0.25 s before and after the join point: the first 0.25 s of the deleted section fades out, and the last 0.25 s fades in, mixing briefly. The result length is therefore exactly what the formula above gives.

• Removing a slip in speech: the default 0.02 s is about right. A long overlap lets part of the deleted words be heard.
• Skipping a bar of music: 0.5 to 1 s sounds natural. The closer to the beat you cut, the less the join shows.
• When little audio is left on a side, or the deleted range is short: the crossfade is shortened so it never exceeds the length remaining on either side or the length deleted; a crossfade longer than the deleted range would play the same sound twice.

MP3 and WAV, and the saved size

MP3 is lossy compression that makes files small by discarding sound you are less likely to hear; WAV stores the sample values as they are. Even if the original is an MP3, this tool re-encodes instead of cutting and joining frames, so it is best to pick a bitrate at least as high as the original. That is why opening an MP3 preselects the closest quality at or above its bitrate, and opening a WAV preselects a WAV with the same bit depth.

One MP3 frame holds 1,152 samples and the encoder emits one extra frame at the end, so the frame count F is ⌈samples ÷ 1,152⌉ + 1. Each frame is ⌊b ÷ sample rate⌋ bytes (b = 144,000 × kbps), and the remainder is made up by adding one padding byte to some frames other than the first. The size is therefore F × ⌊b ÷ sample rate⌋ + ⌈(F − 1) × (b mod sample rate) ÷ sample rate⌉ bytes. The estimated size on screen matches the saved file to the byte. WAV is a 44-byte header plus samples × channels × (bits ÷ 8) bytes, so mono is half the size of stereo of the same length. A constant-bitrate MP3 is the same size in mono and stereo.

When you open a WAV (PCM) file, the sample values are read directly without going through the browser's decoder. So if you save as a WAV with the same bit depth, samples that no fade or join touches keep the same values as the original, and the sample rate is unchanged (16- and 24-bit integer and 32-bit float). An 8-bit WAV is saved as 16-bit and a 32-bit integer WAV as 32-bit float, so those two do not come out byte-identical. Compressed files such as MP3, M4A, FLAC and OGG are decoded at the original sample rate read from the file header, and formats whose sample rate can't be known in advance (such as WebM) are decoded at 48 kHz. When saving as MP3, 32, 44.1 and 48 kHz are kept, and any other rate is changed to a value it divides into evenly (for example 22.05 kHz → 44.1 kHz).

MP3 has a delay that the encoder and decoder add at the start. The MP3 this tool makes does not include the info tag that records that delay, so at 44.1 kHz players add 1,105 samples (about 0.025 s) of silence at the start, and the end is padded to a whole frame. For example, an MP3 saved from a 6.000 s range may show as around 6.034 s in a player. If you need a length that is exact to the sample, save as WAV.

Files you can open, and the limits

One file is open at a time, and any file your browser can decode will open regardless of its extension. Most browsers read MP3, AAC (M4A), WAV, FLAC, Opus and Vorbis (OGG), but the supported codecs depend on the browser and operating system. Even within one extension such as M4A the codec inside can differ, so a file that plays in another app may not open here.

• Up to 30 minutes and 512 MiB: the length is read from the file header, and if a limit is exceeded you are told before decoding.
• Formats whose length can't be read in advance: up to 48 MiB: files with no length in the header, such as WebM or ADTS AAC, are limited by size.
• Files with more than 2 channels: a file such as 5.1 is mixed down to stereo using the Web Audio downmix rules.
• Up to 720 MB once decoded: high-resolution files such as 96 kHz stereo, or 5.1-channel files, are refused above this size even when shorter than 30 minutes.
• Video files: depending on the browser, the sound of an MP4 video may open. If it does not, extract the audio first with an MP4 to MP3 converter.

When you open a file, the whole sound is decoded to 32-bit float and kept in memory. That takes sample rate × channels × 4 bytes per second, and more while the file is being read.

On devices with little memory, such as phones, a long stereo file may not open, or the tab may reload. Close other tabs and apps, and if that does not help, open a file split into shorter pieces.

Common mistakes

• Assuming a low waveform means silence: the waveform height is scaled to the loudest sound in the file. Breathing and background noise can be audible even when they barely show, so check the dB reading in the magnifier.
• Mixing up Keep and Delete: in Delete, the selected part is hatched and is left out of the result. Check once more against the bar and the length in the result area.
• Only renaming the extension: renaming an MP3 to .m4r or .wav does not change what is inside, so it will not work as a ringtone or a WAV. You choose the format when you save.
• Making a fade too long: a 2 s fade on a short spoken phrase buries the first and last syllables. Cut speech at a quiet spot and keep fades to 0.5 s or less.
• Listening but not downloading: 'Play result' and the result player are not saving. The file only lands on your device when you click 'Download'.

Limits and cautions

• Only one range is kept or deleted at a time. To remove several places, reopen the saved result and repeat.
• It does not merge files, record, remove noise, normalize volume, or change speed or pitch.
• Tags such as title and artist, and album art, are not carried into the result.
• Saving as MP3 compresses again, so even an MP3 original loses a bit more. For a lossless cut, save a WAV original as a WAV with the same bit depth.
• It does not save as M4R (iPhone ringtone), AAC or OGG.
• If you reload or close the page, the open file and the result are gone. Download what you need first.

As of The browser documentation, standards and Apple guidance were checked on 2026-10-23. The size tables and example values were calculated with this tool's own formulas, and on the same day we confirmed in Chrome that they match the size of the actual saved files.

Sources

• 01MDN — BaseAudioContext.decodeAudioData()Audio decoded by the browser is resampled to the audio context's sample rate (so the file is decoded in a context at its original sample rate)
• 02MDN — AudioBufferDecoded sound is held in memory as 32-bit floating-point linear PCM (−1 to +1) per channel
• 03W3C — Web Audio API: channel up-mixing and down-mixingThe speaker rules for mixing 5.1 and similar layouts down to stereo
• 04MDN — Web audio codec guideAudio codecs and containers browsers can read, plus MP3 bitrates and sample rates
• 05Microsoft — WAVEFORMATEXWAV format tags (PCM, IEEE float), channels, sample rate and block size
• 06LAME — Technical FAQOne MP3 frame = 1,152 samples, encoder delay and end-frame padding. LAME is LGPL-licensed
• 07wasm-media-encoders — WebAssembly build of the LAME MP3 encoderMIT license (the LAME encoder itself is LGPL). This tool uses version 0.7.0 as is
• 08Apple Support — Create a custom ringtone on your iPhoneA ringtone made in GarageBand can be up to 30 seconds, and it can be shortened automatically if it is longer

Frequently asked questions

Tools to use with it