Audio Splitter
Chop one file into pieces, wherever you want the cuts.
or paste one from your clipboard
How to use it
- Drop the file you want to break up. A two-hour recording opens as fast as a two-minute one, because it is read from disk rather than sent to a server.
- Choose where the cuts go: a fixed number of equal parts, a fixed length for every part, or one cut at each silent gap in the recording.
- Read the summary before you commit. It tells you how many pieces you will get and roughly how long each one is, and it updates as you move the controls.
- Press split. Each piece appears in a list with its own download button, numbered in playing order.
Three ways to divide one file
Splitting audio is the rare editing job where the interesting decision is nothow to cut but where. The cut itself is trivial, copy a range of samples into a new file. So the three methods here differ only in how the boundaries are chosen, and each answers a different question.
Equal parts answers "I need this in n pieces": you fix the count and the durations fall out of it. Fixed length answers "nothing may exceedx": you fix the duration and the count falls out, with a shorter remainder at the end. Silence answers "cut where the recording already has seams", which is the only one of the three that pays attention to the audio rather than the clock.
How silence detection places a cut
The file is scanned in 20-millisecond windows, and the RMS level of each window is compared against your threshold. Twenty milliseconds is short enough to notice a breath and long enough that individual zero crossings in a normal waveform do not register as silence. A run of consecutive quiet windows becomes a candidate gap, and it only counts if it lasts at least the minimum length you set. Which is what stops a two-second pause between speakers from being confused with the tenth-of-a-second beat between words.
The cut goes in the middle of each gap rather than at its start or end. That way both neighbouring parts keep half the pause: the part before it does not end on a hard stop, and the part after it does not start on the first syllable with no run-up. It is a small thing that makes the pieces sound like recordings rather than fragments.
Threshold is the control that matters. It is in dBFS, decibels relative to full scale. So 0 is the maximum a digital file can hold and every real level is negative. Quiet room tone lands near -50, a hissy phone recording nearer -35, a properly noise-floored studio take below -60. If you get no cuts, your threshold is below your room tone; raise it until the count looks plausible.
What each part inherits
Every piece carries the sample rate, bit depth and channel count of the source, because the samples are copied rather than reprocessed. That means the parts reassemble exactly: run them back through the joiner in order and you get the original file, provided you exported losslessly on the way out. Export to MP3 and each part picks up one generation of encoding, the same single generation a normal MP3 export would add, not one per part.
One practical caution: splitting into very many pieces means very many downloads, and browsers throttle those. Around twenty parts is where the process stops being pleasant. If you need a hundred clips out of one file, splitting by length in a few passes is less painful than doing it in one.