A two-hour interview that a transcription service won’t take in one piece. An audiobook that should be one file per chapter. A live recording of a whole concert that you want as separate songs, or a voice memo too large to send. In each case you need to cut one audio file into several, and the question is where to cut and how to avoid damaging the sound.
First decide where to cut
- At the pauses for speech: interviews, lectures, podcasts, audiobooks. Cutting in a silence means no word is split. OpenAI’s own speech-to-text guide, which takes files of up to 25 MB, warns against splitting recordings mid-sentence because the transcription loses context.
- Every N minutes when a service has a length or size limit and you just need pieces that fit.
- In half, or into a set number of equal parts, when you just want fewer minutes per file.
- At chosen points for songs in a concert or DJ set, where only you know where one track ends.
Way 1: online, in the browser
Our Split Audio tool does all four without uploading the file. Drop the MP3, M4A, WAV, FLAC or Opus file and it draws the waveform with the cuts marked:
- Keep At silences to cut in the middle of every pause, or choose Into equal parts
(type 2 to split it in half) or Every… (type
10:00for 10-minute parts). - Check the list of parts. Click the waveform to add a cut, or click a cut to remove one. If a pause is missed, raise the dB number slightly; if a sentence is cut in two, make the shortest silence longer.
- Press Split and download the parts one by one or as a ZIP.
On a 24-second spoken test recording with eight pauses, it found all eight with the default settings and cut each one in the exact middle. The parts keep the original format and are not re-encoded.
Way 2: Audacity, on Windows, Mac or Linux
Audacity is free and splits by labels. Import the file, then add a label wherever a new part should start: place the cursor and press Ctrl + B (Cmd + B on a Mac), or let Analyze > Label Sounds label the stretches between pauses for you. Then choose File > Export Audio, select Multiple Files, set Split files based on to Labels, tick Include audio before first label if the start of the file should be a part too, and export.
One thing to know: Audacity decodes everything you import, so exporting an MP3 encodes it again. For speech at a decent bitrate you are unlikely to hear the difference after one round, but it does add up if a file goes through several edits. Export as WAV or FLAC if you plan to edit the parts further.
Way 3: FFmpeg, from the command line
Search for “split audio ffmpeg” and you will find this command, which cuts a file into 5-minute (300-second) parts without re-encoding:
ffmpeg -i input.mp3 -f segment -segment_time 300 -c copy part%03d.mp3It is fast, and for most uses it is fine. But we wanted to know if the parts really add up to the original, so we cut our test MP3 into 5-second parts with it, decoded them and joined them end to end. The joined parts were 129 milliseconds shorter than the original. Lining each part up against the original showed where the sound went: about 12 ms was missing at each cut, and 93 ms at the last one.
The reason is how MP3 decoding works. Every MP3 file starts with a short delay, 529 samples, that decoders produce and then throw away, along with any padding the encoder noted in the file’s header. At the start of the original file those samples really were padding. Each part, though, is a new MP3 file, so its first 529 samples are thrown away too, and in a part cut from the middle they are real sound: 529 samples is exactly the 12 ms we found missing at each cut. For a podcast cut in a pause that doesn’t matter. For music split into tracks that should play back to back, it is a small but audible jump.
Without -c copy, FFmpeg decodes and re-encodes each part instead. The parts then join without a gap, but
the sound has been compressed a second time.
What we measured
| Method (24-second MP3) | Re-encoded | Joined parts vs original |
|---|---|---|
FFmpeg -f segment -c copy, 5-second parts | No | 129 ms shorter: ~12 ms lost per cut |
FFmpeg -ss / -to per part, default settings | Yes | Same length; compressed a second time |
| Split Audio, 9 parts at the pauses | No | Identical, sample for sample |
Our tool avoids the gap by doing two things the FFmpeg recipe doesn’t: each MP3 part starts two frames before its cut, so the decoder has what it needs for the first note, and its header gets a new, correct number of samples to skip, so players start exactly at the cut and stop exactly at the next one. We checked the joined parts with two decoders, Chrome’s and FFmpeg’s, and both matched the original sample for sample. M4A, WAV and FLAC parts matched too.
On a phone
The Split Audio page works in phone browsers as well: open the file from Files (iPhone) or your downloads (Android). An iPhone voice memo is usually an M4A file with AAC sound and splits without re-encoding; memos recorded with the Lossless setting need the WAV option. If you only need one section rather than several parts, Trim Audio is simpler.
Quick answers
How do I split an audio file in half? Choose Into equal parts and type 2. The cut lands at the exact middle; if you would rather cut at the nearest pause, use At silences and remove the other cuts.
Does splitting an MP3 lower the quality? Not if the frames are copied, as with -c copy or
our tool. Programs that open the file, edit it and save a new MP3 compress it again.
How do I put the parts back together? Merge MP3 joins MP3 files without re-encoding. Parts made with our tool join back into the original sound with no gap.