Interval Ear Trainer

Interval recognition is one of the most practical skills in ear training: it is how a bassist finds a root by ear, how a singer nails a harmony part without sheet music, and how a producer transposes a hook they can only half-remember. Unlike absolute pitch, relative interval recognition is a trainable skill — the ear learns to recognize the distance between two notes as a shape, and that shape stays the same no matter what key it happens to land in. This tool is a focused, repeatable drill for exactly that skill: hear two notes, name the gap between them, and get instant feedback.

Every round is generated on the spot with the Web Audio API — two OscillatorNode instances, each running through its own GainNode envelope so the attack and release ramp smoothly instead of clicking at the start or end of a note. The root note lands on a random MIDI pitch within about an octave of middle C, and the second note is computed by simple semitone math (frequency = 440 × 2^((MIDI−69)/12)), so the exact same interval never sounds identical twice — the register shifts every round, which is exactly what forces the ear to recognize the interval itself rather than memorizing two specific pitches. A toggle switches between melodic playback (the two notes play one after another, the classic ear-training format) and harmonic playback (both notes ring together, closer to how intervals are actually heard inside a chord).

The practice session adapts as you go. Behind the scenes, every interval carries a running correct/wrong tally for the session, and each time a new question is generated, the pool of 13 possible intervals — unison and the octave bracketing the eleven semitone steps from minor 2nd through major 7th — is drawn from a weighted random pick rather than a flat one. An interval you keep getting wrong gets a higher weight and shows up more often; one you have nailed several times in a row settles back toward the baseline. Before any answers are logged every interval starts at the same weight, so early rounds are genuinely uniform — the bias only appears once real mistakes give it something to react to. The accuracy table beneath the quiz shows the exact correct/wrong counts and percentage behind every interval, so the weighting is never a mystery.

Because everything — synthesis, timing, scoring, and the adaptive weighting — runs locally in the browser, there is nothing to install, no microphone permission to grant, and no account or login. The score and per-interval stats live only for the current session; refreshing the page starts a clean slate. Treat a quiet run of correct answers on an interval as a genuine signal that it has settled into your ear, and treat a persistent trouble spot the same way a private teacher would: as the interval worth the next ten minutes of focused listening.