1

What can be recommended as challenging tests for cross-fading/mixing of music?

I am looking for examples of music-titles that exhibit severe volume dips when cross-faded or mixed.

Ideally answers should provide pairs of (free) music-titles and the length of the cross-fading interval and also other information that is necessary to exhibit the volume dips.

Manfred Weis
  • 133
  • 4
  • 2
    All kinds of musical (or not) pieces can give bad crossfade results if not done properly. – ZaellixA Feb 23 '23 at 10:06
  • 1
    Such examples would be purely artificial. You can generate two sound files that when summed up together at correct sample offset and correct crossfade coefficients, will cancel each other out to zero value. Just generate two pure sine waves and invert the other signal or change the time offset so the two waves are offset by 180 degrees. – Justme Feb 23 '23 at 10:48
  • 1
    @Justme I can imagine music that, to the ear, cross-fades poorly; 50% volume doesn't affect all sound equally, similar to colors and brightness to eyes. It does depend what OP means by "volume dips" but that's one meaningful take. – OverLordGoldDragon Feb 23 '23 at 11:44
  • @ZaellixA true, but I would like examples that sound bad when cross-faded or mixed even if done by a renowned sound engineer who gives his/her best. – Manfred Weis Feb 23 '23 at 16:05
  • @Justme why would these examples be purely artificial? I have of course already tested all kinds of artificial data so I now want to demonstrate the importance of the issue to lay people that only care for good sounding cross-fadings and mixings. – Manfred Weis Feb 23 '23 at 16:10
  • @OverLordGoldDragon the volume dips are common term used for describing the effect that cross-fading equally loud audio files results in a gradual decrease of volume towards the middle of the fading interval followed by a gradual increase twoards the end. Take the crossfading of sin(2000x) with -sin(2000x), i.e. x(sin(2000x)-(1-x)sin(2000x) and you'll see the effect. – Manfred Weis Feb 23 '23 at 16:19
  • 1
    @ManfredWeis That's what I meant. If you play the two sine waves out of phase, they start to cancel each other. You are summing a sine wave with it's exact inverse, which will output zero at 50% mix. If you play them exactly in phase, you cannot see any difference. That's why I said any sound files exhibiting the effect are purely artificial. No real piece of sound will do that. They are just cancelling sine waves, not two equally loud sound files like you originally requested. – Justme Feb 23 '23 at 16:41
  • @ManfredWeis if you really need help I guess you would have to define "sounds bad" in a more quantitative way. I've yet to experience a bad cross-fade when done by well trained sound engineers with good monitoring systems. – ZaellixA Feb 23 '23 at 18:58

2 Answers2

1

Crossfades definitely have a "sweet spot" in terms of cross fade length. Too short, it sounds more like artifact and too long: you listen to both songs at the same time which is often weird especially if they clash in time signature and key.

Most difficult tend to be songs that start or stop with a big bang or have a slow fade or fade out. An extreme example is for a slow start and big bang ending is Maurice Ravel's "Bolero".

So difficult song pairs would have

  1. Clashing keys (e.g. half step apart) especially if they also have long chords in the intro and outro
  2. Clashing timing (speed difference by maybe 10%-20%)
  3. Slow fade out into a slow fade in (ends up with a big hole in the middle)
  4. Big bang start (starting chord gets cut off)
  5. Big bang end (ending chord gets cut off)
  6. Any combination of the above

Crossfades are generally not a good idea and I avoid them whenever possible.

Hilmar
  • 44,604
  • 1
  • 32
  • 63
  • 2
    Sometimes Hil, there is no sweet spot. That is why time-domain pitch shifters and tempo changers sometimes sound like shit and frequency-domain algs like phase vocoder and sinusoidal modeling sound better. They're both splicing and crossfading. And the splice displacement is meant to be an integer number of cycles or periods of similarity. It's just that the latter is doing in on a per frequency component basis, so the splice is always glitch free. All the time-domain methods can do is mitigate glitches. – robert bristow-johnson Feb 23 '23 at 22:40
  • @robertbristow-johnson your hinting to the phase vocoder algorithm is the most useful information I got from this thread; still hoping to get what I actually asked for besides the dos and donts – Manfred Weis Feb 24 '23 at 05:32
  • @robertbristow-johnson If you could turn your comment into an answer, elaborating a little bit on phase vocoding and sinusoidal modeling, I would accept it because these two things actually solve the issue of volume dips. – Manfred Weis Feb 24 '23 at 06:52
  • Well @ManfredWeis, the equal-power crossfade can also solve the problem of volume dips. Using a phase vocoder to morph from one sound file to another is a more complicated thing. – robert bristow-johnson Feb 24 '23 at 17:24
1

Perhaps some cross fading difficulties are best understood in terms of musical terms (key, tempo) while other difficulties are best understood in dsp terms (phase cancellation)?

I like the comments from others about «morphing» from one track to the other. Ideally one might (?) like one track to seamlessly morph into another based on perceptually relevant parameters (tempo1->tempo2, key1->key2, piano into saxophone, …). Compared to that, a simple volume ramp up/down mix seems crude.

Knut Inge
  • 3,384
  • 1
  • 8
  • 13