Nightcore. And the sped-up version.

Nightcore and "sped up" are the same effect at different strengths

Nightcore started as a pair of Norwegian DJs pitching trance up to make it harder and brighter, and the name stuck to the technique rather than the genre. What it does is simple and specific: play the track faster, and let the pitch rise with the speed, exactly as a record does. The vocal turns light and young, the hi-hats get glassy, and the energy goes up because everything is genuinely arriving sooner.

The wave of "sped up" edits on TikTok and streaming services is the same operation with the dial turned down. Around 1.15× to 1.20× the effect reads as energy rather than as a transformation, and it is quiet enough that many listeners never register it as an edit at all. Above about 1.30× nobody could miss it.

The pitch column is not an estimate. It is twelve times the base-two logarithm of the speed, so 1.25× is 3.86 semitones — a little under a major third — and 1.335× would be exactly a fourth.

Why upward speed is less forgiving than downward

A slowed edit hides its artefacts. A sped-up one advertises them, for two reasons.

Formants move with the pitch. The resonances of a human vocal tract tell your ear roughly how big the speaker is, and they do not move when someone sings higher — they are a property of the body, not the note. Raising the whole signal raises them too, so the singer appears to shrink. Two or three semitones reads as bright; five semitones reads as a chipmunk. This is a real limitation of the technique rather than a fault in this implementation, and no amount of processing here can separate the two, which is why the speed list stops at 1.35×.

Everything shifts up into the range your ear is most sensitive to. Sibilance that sat politely at 7 kHz lands near 9 kHz at 1.25×, where the ear is more critical. Harshness in the original comes back doubled. If the result is fatiguing rather than bright, a small cut around 8 kHz in the EQ tool does more for it than a lower speed would.

There is also a practical trap: the top end runs out. A 128 kbps MP3 is already cut off around 16 kHz, and speeding it up pushes that ceiling to about 20 kHz, which is fine — but the encoder's artefacts move up with it and become easier to hear. Starting from a lossless file and choosing WAV output avoids stacking a second generation of compression on top.

Reverb, sparingly

The reverb here is a convolution reverb built from an impulse response the page synthesises: decaying noise, independent per channel so the tail spreads across the stereo image rather than collapsing to the middle, with a short pre-delay before the tail starts and a low-pass on the wet path so it darkens as it decays the way a real room does.

All of that matters much more on the slowed page, where the long tail is the point. On a fast track a long tail is just mud: the next transient arrives before the previous one has decayed, and the mix loses its edges. The default here is a short, quiet room that takes the hardness off a bright vocal, and "none" is a perfectly normal choice — most nightcore is dry.

Because the wet signal is added to the dry rather than crossfaded with it, the output peak sits above the input peak. On a mastered track the standard setting measured 1.05 before correction, which would clip. The render is scaled back when that happens and left untouched when it does not, so a quiet recording keeps its level instead of being normalised up.

Practical notes

The classic nightcore workflow is two steps. Traditionally the vocal is lifted from one track and laid over a faster instrumental. If that is what you are doing, the stem splitter separates vocal from instrumental with a neural model running on your own machine, and the joiner puts pieces back together.

Tempo matters if you are mixing. Speeding a 128 BPM track to 1.25× gives 160 BPM exactly. If you are matching it to something else, the BPM finder reads the tempo of the original so you can pick a speed that lands on a round number rather than discovering afterwards that you are at 159.4.

Batch works. Several files at once come back as a zip, all treated identically. The effect itself is fast; the encode is the slow part.

Tags do not survive. The output is a new file, so title and artist are not carried across. The tag editor writes them back without touching the audio, so it costs no extra compression.

On posting the result. A sped-up edit of a commercial recording is still that recording, and content matching is not defeated by a speed change — it has not been for years. Expect a public upload to be identified. What that means for you depends on the platform and on whether money is involved; it is worth knowing before rather than after.