Why is my correlation jumpy on a steady signal?

The meter computes correlation on a short 2048-sample window (~46 ms at 44.1 kHz) updating ~30 fps. Steady periodic signals hold near-constant correlation; complex material with transients jumps naturally as the spectrum changes window-to-window.

🔄

Audio Phase Alignment / Inversion Tool

A live stereo phase correlation meter (−1 to +1) with polarity invert, millisecond-accurate delay alignment, vectorscope, and per-channel waveform display. Use a built-in test signal to feel how delay and polarity affect correlation, or upload your own stereo file to check it.

Source

Test waveform

Frequency 440 Hz

Same signal feeds both channels — the only difference between L and R is the delay & polarity you set below.

Right-channel processing

Delay R relative to L 0.00 ms

0–50 ms range. 1 ms ≈ 44 samples at 44.1 kHz, or ≈ 0.34 m of air path.

Invert R polarity (×−1)

Volume 25%

Phase correlation

+0.000

Uncorrelated (~stereo)

−1 anti-phase 0 uncorrelated +1 mono

Vectorscope (L horizontal, R vertical)

dots = sample pairs — — diagonals = ±1 reference

Reading the Vectorscope & Correlation Meter

The phase correlation meter is a number between −1 and +1 that tells you, in a single glance, how much your two stereo channels resemble each other in time. The vectorscope visualises the same information as a 2-D plot: every audio sample is one dot, plotted with the left channel's amplitude on the X axis and the right's on the Y axis.

What the readings mean

+1.0 — both channels are identical. On the vectorscope: a 45° line from bottom-left to top-right. This is what mono sounds like through stereo speakers.
+0.5 to +0.9 — heavily correlated, "mostly mono" (vocals + reverb, a centred kick drum).
0 — uncorrelated (truly stereo material, decorrelated reverb, two unrelated mics). On the scope: a rough cloud or disc.
−0.5 to −0.9 — the channels are anti-phase. Partial cancellation when summed to mono. Symptom of a polarity flip, a misrouted speaker cable, or a serious time alignment problem.
−1.0 — perfect anti-phase. On the scope: a 45° line from top-left to bottom-right. Mono summing produces silence.

The mono-compatibility test

Any time you have stereo content with a negative correlation across the lows, mono playback (a phone speaker, a small radio, a club PA bridged-to-mono) will eat that energy. Sub-bass and kick drum in particular need to live near the centre with strong positive correlation; if your vectorscope is bouncing all over the place in the low end, something's misaligned.

How to use this tool

Start in Test signal mode at 440 Hz sine. Hit Play — the meter sits at +1.0 (identical L and R). Slide the delay up: as it crosses about 1.1 ms, you'll see correlation drop through 0 toward −1. That's the period of 440 Hz divided by 2 — a half-cycle delay puts the channels exactly out of phase. Toggle Invert R polarity and watch the correlation flip sign instantly. This is the same effect as flipping a speaker's wire polarity or hitting the phase switch on a mic preamp.

Real material

Switch to Upload audio file and load a stereo mix. Play it back — the correlation will bounce around as the music moves. Healthy modern productions sit between roughly +0.3 and +0.8 most of the time; rock and electronic often stay close to +0.6 to +0.9 (centre-heavy mixes); jazz / classical / ambient may dip down to +0.1 or below (wider stereo image). What you don't want to see is sustained excursions below 0 — that flags a real phase problem.

Frequently Asked Questions

My drum overhead mics show correlation around 0. Is that a problem?

Probably not. Overheads usually capture wide stereo content (a spaced-pair or ORTF arrangement decorrelates by design), and a correlation near 0 is exactly what you'd expect. The warning signs are sustained negative correlation (especially on the kick or snare hit) — that indicates a polarity or time issue between the two overheads or between overheads and the close mics.

What's the difference between polarity inversion and a 180° phase shift?

A polarity invert is an instantaneous sign flip: every sample is multiplied by −1, at every frequency. A "180° phase shift" depends on frequency — a constant time delay shifts a 100 Hz signal by 1 cycle (no audible effect) but a 200 Hz signal by 2 cycles, etc. Loosely, audio engineers often say "phase invert" when they mean "polarity invert"; this tool's checkbox is a true polarity invert.

Why does the test signal show negative correlation at certain delays?

At a delay equal to half the test signal's period, the two channels are exactly out of phase: the right channel is at its peak when the left is at its trough. For 440 Hz, half-period is 1/880 ≈ 1.14 ms. At 100 Hz it's 5 ms; at 1 kHz it's 0.5 ms. Slide the delay up while watching the meter — every time you cross a half-period you cross through −1, and every time you cross a full period you cross back through +1.

How accurate is the delay setting?

The DelayNode in Web Audio is sub-sample-accurate (the API uses linear interpolation between samples). The slider is in milliseconds with 0.1 ms steps — at 44.1 kHz that's about 4.4 samples per step. For mic-to-mic alignment work you typically need 0.01–0.5 ms precision; this tool covers that with the 0.1 ms slider step.

Can I align my drum overheads and kick mic with this?

In principle yes — load a stereo "kick close mic on L, overhead on R" file and slide the delay until correlation peaks on the kick hits. In practice it's faster and more precise to do this in your DAW with sample-accurate tools (Auto-Align, Track Spacer, plain nudging). This tool's value is showing you the concept and giving a quick correlation check on any stereo content.

Is my file uploaded anywhere?

No. The file is read into the browser via FileReader and decoded with the Web Audio API — both run entirely on your machine. Nothing is uploaded, nothing leaves your browser. There's no analytics on the audio content.

Why is my correlation jumpy even on a steady signal?

The meter is computed on a short 2048-sample window (≈ 46 ms at 44.1 kHz) that updates ~30 times per second. Steady periodic signals (sine, sawtooth) hold a near-constant correlation; complex material with transients (drums, music) naturally jumps as the spectrum changes from window to window. That's normal.

Related Tools

← All Signal Processing Tools