i
Open a song or a recording, cut the piece you want, shape the volume, and save
it as MP3 or WAV. Everything happens in this tab; nothing is uploaded.
Every online audio cutter and converter uploads your file to somebody's
server first. This one has no endpoint: turn off your wifi after it loads and
it still works.
1. Open a file
Drop audio here, or click to chooseMP3, WAV, OGG, FLAC or M4A. Several at once go to the playlist. Nothing is uploaded
or record instead of opening a file: it lands in the playlist, ready to trim and save as MP3
0 kbps0 khzmonostereo
no file loaded
2. Cut and shape it
+0 dBas loud as it can be without clipping
s
s
changes the pitch with it, the way tape does
cancels what both channels share, which in most mixes is the lead vocal
3. Save it
4. The playlist
Open more files and they line up here. Join stitches them into one file, in
this order, each one whole. Click a row to put it in the player.
A song or a recording in; a trimmed, louder, faded or joined one out, as MP3 or WAV.
The face is drawn after the player everyone actually had in 1999, and the jobs
are the ones people still search for every day: cut a song down to a ringtone,
convert a recording to MP3, make a quiet file louder, fade the ends, stitch
two recordings together, record straight from the microphone, pull the vocals
out of a song. Each one is a couple of clicks, and the before and after never
leave this tab.
Is this actually private?
Nothing is uploaded. Static page, no endpoint, no account. The file is decoded in this tab and the result is written back out of it.
Decoding is the browser's, so what opens depends on it. MP3, WAV, OGG, FLAC and M4A open everywhere current; WMA and AIFF mostly do not, and the error will say what the file was. Everything is decoded at the machine's own rate, usually 44.1 or 48 kHz, which is also what gets written back out.
The MP3 encoder is LAME. The real one, as the lamejs port, about 165 KB served from this domain the first time you save, and pinned byte-for-byte by a test in the repository. Writing our own psychoacoustic model would be exactly the kind of reckless this site exists to avoid.
WAV out is 16-bit PCM; MP3 out is lossy, and re-encoding an MP3 costs a generation. The WAV is lossless for anything that started at 16 bits, which is nearly everything; a 24-bit or float master is reduced to 16 bits once, on the way out. Trimming an MP3, by contrast, decodes and re-encodes it, so the result is a copy of a copy, inaudible at 192 kbps and up, but if you plan to edit further, save WAV and encode to MP3 once at the end.
No Opus or AAC out, on purpose. The browser can only record those in real time, so a thirty-minute file would take thirty minutes to save. A tool that quietly did that would be wasting your afternoon to look modern; MP3 and WAV cover what people actually come for.
Normalise is peak, not loudness. It scales the file so its loudest sample sits just under full scale. Proper loudness matching (LUFS) is a different, heavier measurement, and calling a peak scale that would be a lie. Peak is still what most people reaching for the word want: as loud as it goes without clipping.
Speed changes pitch. 1.5x here is a tape played fast: shorter and higher. Stretching time while keeping pitch is real signal surgery with real artefacts, and this page does not pretend to it.
The join takes each file whole, in playlist order, resampled to one rate, with a mono file lent to both sides of a stereo one. The cut-and-shape settings apply to the single file in the player, not to the join.
Whole files sit decoded in memory. Decoded audio is about 20 MB a minute in stereo, so songs are nothing and a three-hour recording is real memory. If a tab dies on a huge file, that is where it went.
Tags do not survive. A decoded buffer has no idea what album it came from, so the output carries no ID3 tags. What you keep is the sound.
The recorder uses your browser's own encoder. Record asks for the microphone, and only then: the permission prompt is the browser's, the sound goes into this tab and nowhere else, and stopping drops the recording into the playlist decoded, ready to trim and save as MP3 or WAV like anything else. Nothing records without the red dot showing.
The vocal remover is centre cancellation, stated plainly. Most mixes pan the lead vocal identically into both channels; subtracting one channel from the other cancels whatever the two share. That is the whole trick: it needs a stereo file, it takes bass and kick with the vocal unless "keep the bass" mixes the low end back, and a vocal with a wide stereo reverb leaves its reverb behind. On a typical mix it works surprisingly well; on some it barely works at all, and no tool that runs on arithmetic can promise otherwise.
The spectrum is decoration with a pedigree. It is a real analyser on the playing audio, drawn the way the original drew it, and it does nothing to the file. With animations reduced in your system settings it stays dark.
Nothing here is for sale, nothing is tracked, and there is no account to make.
If something saved you an hour: 83TQcTwusSQ4WKbPQE5osrF3cR4GWe2zmcNWeozK6BSqHSaeLvjUVe476ouVwLKn1uVwEFcbJQvnme7W6dTV5SB93x45DEy
You can
check that address here before sending anything.