The one number that says your low end will survive mono

Correlation is a measurement, checking in mono is a habit

Every article about a mix that falls apart in a car says the same thing: check it in mono. It is good advice and it is not a measurement. It tells you to go and listen, it does not tell you what you are listening for, and it gives you nothing you can write down and compare against the next bounce.

Correlation is the number underneath that advice. Take the left channel and the right channel as two signals and ask how much they agree. The answer runs from 1 to minus 1. At 1 the two channels are the same signal, which is what mono means. At 0 they are unrelated. Below 0 they are actively cancelling, and summing them removes energy that you can currently hear.

For low end, one number matters more than the full band one, and it is the correlation of everything below 150 Hz.

Key takeaways

• Low band correlation is the correlation of the two channels after both have been filtered to keep only what is under 150 Hz.
• The value is between 1 and minus 1. Near 1 the bass is effectively mono and survives being summed. Near 0 it is wide. Below 0 part of it disappears when summed.
• We call a mix low end mono compatible at 0.6 and above. That line is a choice we made, not a number we calibrated against a labelled corpus, and the honest thing is to say so.
• The number does not tell you which element is wide. It tells you that something is.
• Full band correlation and low band correlation are different readings. A mix can be wide and healthy at 0.42 overall while its bass sits at 0.95.

How the number is computed

There is no clever part, and that is deliberate. The steps are short enough to argue with.

1. Take the left and right sample arrays as they are.
2. Compute the Pearson correlation of the two. That is the full band number.
3. Compute the side channel, which is left minus right, halved, and take its level in dBFS. That is a width reading, not a correlation.
4. Filter both channels through a fourth order low pass at 150 Hz, built as two biquads in series, so only the bass band survives.
5. Compute the Pearson correlation of those two filtered signals. That is the low band number.
6. Round the correlations to two decimals and the side level to one.

A mono file skips all of it and returns 1.0 by definition, with the side energy floored, because there is no second channel to disagree with.

The line at 0.6 is a choice, and here is what that means

We had to pick a value at which the tool stops saying nothing and starts saying something. We picked 0.6. It is in one place in the code and it is one comparison.

We did not derive it from a labelled corpus of mixes that were known to translate and mixes that were known to collapse, because we do not have such a corpus. We are not going to describe it as validated, tuned or trained. It is a gate, it is documented, and a reader who thinks it belongs at 0.5 or 0.75 is having a reasonable disagreement rather than catching us in an error.

What we can say about it is narrower and true. Above the line, summing the bass to mono changes very little, because the two channels were already carrying nearly the same low frequency signal. Below it, summing removes energy that is currently audible in stereo. How much it removes depends on how far below the line you are, and 0.12 and 0.55 are not the same problem even though the gate treats them the same way.

That is why the number is reported and not only the verdict. A verdict alone is a threshold pretending to be a measurement, which is the same failure we wrote about when an analyzer says your mix is fine and it is not: level thresholds looked clean because they were measuring the wrong thing.

Why headphones hide this specific fault

On headphones each ear receives one channel and only that channel. The two signals never meet in the air, so the cancellation that happens when a mono system sums them never happens. A bass that is spread wide sounds larger on headphones for exactly the reason it will sound thinner on a system that sums.

That system is not rare. Club systems commonly sum the bass below a crossover point on purpose, because two low frequency drivers fed different signals produce a room dependent mess. Car systems, phone speakers and single speaker bluetooth boxes sum for simpler reasons.

So headphones do not lie to you in general. They lie to you about this.

What the number cannot tell you

Four limits, all of them real.

It does not name the source. A correlation of 0.2 under 150 Hz says the bass band disagrees between channels. It does not say whether that is a wide sub, a stereo reverb on a kick, a doubled bass part panned apart, or a saturator with a stereo mode you forgot was on. You get a fact, not a diagnosis.

150 Hz is a boundary we drew. Low end does not stop there. The filter is fourth order, so content just above 150 Hz is attenuated rather than removed. Move the boundary and the number moves.

Correlation is not phase. A signal and the same signal delayed will correlate poorly, and so will a signal and a different signal. The reading cannot separate a timing problem from an arrangement problem.

It says nothing about level. A mix can be perfectly mono below 150 Hz and still have far too much energy there, or almost none.

What is a good low band correlation value?

Above 0.6 by our gate. Closer to 1 means the bass is closer to mono. There is no advantage to being wide down there on a system that sums, and there is a cost.

Does making my bass mono make the mix smaller?

Below roughly 100 to 150 Hz, very little of what you hear as width lives there. Width in a mix is mostly carried above that. Collapsing the bottom usually costs less than people expect and buys back level, because a wide bottom drives a limiter harder for the same perceived weight.

Why is my full band correlation low but my bass fine?

Because they are separate readings. A wide stereo image with a mono bass is a common and healthy shape. Judging the low end by the full band number is how a correct mix gets flagged.

Can I measure this without a plugin?

Yes. Any correlation meter reads the full band number. The low band number requires filtering both channels first, which is why it is less commonly displayed even though it is the one that predicts translation.

Limits

The 0.6 gate is not corpus validated, as stated above. The claim that low frequency width costs limiter headroom is physics we act on rather than a run we performed. A tool that reads your session metadata instead of your audio cannot produce this number at all, which is the line the category splits on. And every number here is computed from the file or the live audio, so a measurement taken from a loop that has not reached the wide part of the arrangement will describe that loop and nothing more.