Polarity ABX Test — Phase Drill

A blind polarity test evaluates whether a listener can distinguish between normal and inverted acoustic phase on musical material. Inverting polarity flips the waveform upside down, testing if the ear perceives transient attack direction, bass firmness, and spatial focus differences when positive pressure waves become negative pressure rarefactions. This drill uses an asymmetric 220 Hz pulse, the class of waveform where the question is genuinely open, and scores your answers against pure chance.

How to use it

1. Fit closed-back headphones at conversational volume. The 220 Hz pulse needs to be felt as much as heard, but the session runs minutes long, so keep it light.
2. Learn the references. Signal A plays the pulse in normal polarity; Signal B plays the same pulse with every sample sign-flipped. A/B buttons let you audition them openly, as many times as you like.
3. Take the blind trials. The engine plays X, randomly drawn from A or B, and you answer which reference it matched. There is no way to know the assignment in advance; the button order is shuffled and the labels carry no state.
4. Watch the statistics build. After each answer the panel updates trials, correct count, running p-value, and your consecutive-correct streak.
5. Accept the verdict only when significance lands. The badge reads significant, needs more trials, or chance based on the binomial math below, not on how confident you feel.

Polarity versus phase, and why asymmetric waveforms matter

Polarity inversion is the mathematical operation x(t) → −x(t): every sample multiplied by −1, equivalent to a 180° shift at every frequency simultaneously, with no delay added. A genuine phase rotation shifts one frequency band relative to another and changes the spectrum’s timing, not its sign. The two are routinely conflated, but they are different operations with different audible consequences.

Here is the key fact that makes this a real test. Flipping the sign of a signal changes no magnitude anywhere: |−X(f)| = |X(f)|, the FFT amplitudes are bit-for-bit identical, only the phase terms rotate by 180°. A perfectly linear, symmetric detector therefore cannot distinguish the two, which is why polarity inversion on a pure sine is provably inaudible in isolation. The question stays open only for asymmetric waveforms, where the positive peak and negative peak differ in size or shape. A kick drum’s initial beater attack, a brass note’s pressure onset, and voiced speech all compress air outward first and harder than they rarefact. This tool’s 220 Hz stimulus is built the same way: an asymmetric pulse whose “push” orientation is the only thing that changes between A and B.

Two physical asymmetries can convert that waveform difference into a perception. The transduction chain of the inner ear behaves partly like an acoustic half-wave rectifier, a mechanical nonlinearity documented in the classical AES phase-audibility literature, so a compression-led attack and a rarefaction-led attack do not stimulate the cochlea identically. And loudspeaker cones have asymmetric suspensions: the same voltage drives the cone out more easily than in, a fact behind the standard wiring convention where positive voltage on the red terminal pushes the cone outward. The published listening data is humbling either way: in the exchanges between Shanefield and Lipshitz’s group in the JAES correspondence, the parties converged on absolute polarity being “extremely subtle” for most listeners, more detectable below about 500 Hz, and rarely preferred once detected. This drill asks what you hear, under conditions where guessing cannot hide.

How the double-blind protocol controls bias

The ABX design exists to defeat the three ways listeners fool themselves about phase: expectation (“inverted should sound thinner, so it does”), order memory, and the shift in criterion when a listener knows which button is “correct.”

Signal A and Signal B are fixed and openly labeled. Signal X is assigned by a fresh random draw each trial. The page shows you nothing about the draw. Your only task is match-or-not, and your running score is compared against the null hypothesis of coin-flip guessing using the upper-tail cumulative binomial probability:

P(K ≥ k out of n trials) = Σᵢ₌ₖₙ C(n, i) × 0.5ⁿ

The table below is the actual decision boundary the engine works against at the α = 0.05 criterion standard in subjective testing.

Trials	Correct answers	p-value	Verdict
5	5	0.0313	significant
8	7	0.0352	significant
9	8	0.0195	significant
10	9	0.0107	significant
15	12	0.0176	significant
10	7	0.1719	not significant

Five perfect picks already beats chance, but 7-of-10 does not: that pattern arises by luck about one session in six. This is why the engine refuses to grant a verdict on shaky evidence and keeps the needs more trials badge up instead. The α = 0.05 convention — tolerating a one-in-twenty false positive — is the same statistical discipline used by formal subjective frameworks such as ITU-R BS.1116 for small-impairment assessment, adapted here for a single listener at a browser.

Accuracy, limits and troubleshooting

• You are testing one orientation per session. If you score significant, re-run with the physical roles swapped, or check on a second track; a single pass proves perception, not preference. Discriminating a difference and liking it are different claims, and the literature notes listeners rarely prefer either polarity.
• One speaker inverted is not this test. Flipping a single channel of a stereo pair causes destructive summing in the bass (at 100 Hz, wavelength is 3.43 m, so room path differences decide the comb pattern), thin lower midrange, and a drifting phantom center. That is a wiring fault; the Speaker Test checks it directly.
• Disable OS audio effects. Loudness equalization, spatial audio, and “bass preservation” modes resample and filter the pulse, altering its asymmetry.
• If A and B sound identical, that is a legitimate result. Most listeners on most material hear nothing; the drill’s value is measuring it rather than arguing about it.

Related tools

• Blind ABX hub — the full threshold series with shared scoring.
• Speaker Test — verify left/right wiring polarity with cancellation tones.
• Tone Generator — build your own asymmetric references with summed harmonics.
• Headphone Test — confirm your headphone passes both channels cleanly before trusting a session.