R and L: what does the recognizer actually hear when I mix them up?

It hears two genuinely different tongue shapes and reports which one your recording actually matched, which is precisely why this pair is worth checking with a tool rather than your own ear alone.

30-second version: English /l/ touches the tongue tip to the ridge behind your top teeth; English /ɹ/ (the "r" sound) never touches anywhere — it curls or bunches back without contact while the lips round slightly. Several East Asian languages have one flap sound that sits near both, so the two English targets can genuinely blur for a learner in a way that's about tongue shape, not effort.

Why this one is unusually hard

Most consonant confusions come from one language having a sound the other lacks. R/L is trickier because the "in-between" sound — a quick tongue tap, common in Japanese, Korean, and several other languages — resembles neither English target very closely once you compare tongue shapes carefully. English /l/ is a full contact sound: tip on the ridge, air flows around the sides. English /ɹ/ is a no-contact sound: the tongue approaches but never touches, while the lips round like the start of a /w/. A single tap that briefly touches the ridge sits closer to /l/, so an untrained ear reaching for something more r-like tends to either keep tapping (heard as /l/) or drop the tap and round the lips too much (drifting toward /w/, a different swap covered in the v/w problem).

Because the difference is contact vs no-contact rather than a subtle vowel shade, it's one of the more mechanically checkable contrasts once you know what to measure.

What the recognizer measures

sayit's phoneme model doesn't classify "r-ish" or "l-ish" impressionistically — it reads the acoustic signature each sound leaves in the audio. /l/ has a distinctive resonance pattern from the tongue-tip contact and the air flowing around the sides; /ɹ/ has a characteristic drop in the third formant from the tongue curling back with no contact at all. Those are physically different signals, not degrees of the same sound, so the model's confidence separates them even in cases where a human listener, trained on a different sound system, might genuinely struggle.

When you say "right," the target is /ɹ aɪ t/. If the acoustic evidence matches /l/ instead, the breakdown shows a substitution at that first slot — and because "light" is a real word, the "sounded like" hint will usually name it directly.

What the fix looks like

For /l/: touch the tongue tip firmly to the ridge just behind your top teeth, and let air escape around the sides — you should be able to hum /l/ continuously with the tip planted.

For /ɹ/: the tongue never touches anything. Pull the tip back and slightly up without contact, and round the lips a little, almost like starting a /w/. If your tongue touches the roof of your mouth anywhere during the sound, you've made an /l/ or a tap instead.

A useful diagnostic: say a long, held "rrrrr" and a long, held "llll" back to back and notice whether your tongue tip moves. If it stays still for both, one of them isn't landing.

Drill it

Start with right vs light and rice vs lice, then move to road vs load and rock vs lock once the first pair is reliable. The /phonemes/r and /phonemes/l pages walk through the tongue position in more detail than fits here.

Record: "The rally leaders really loved the local river." It forces both sounds back to back four times, which is where the habit either holds or slips.

How do I know it's improving?

Check whether the substitution direction is consistent across the drill sentence and a fresh, unrehearsed sentence you haven't practiced. Progress on a memorized drill sentence alone can reflect motor memory of that specific sentence rather than the sound itself; the real test is whether a new sentence with the same contrast gets a clean verdict too.

Try it

Open sayit free — no install, no card — and read the trap sentence above once cold, before you've warmed up. That first, unrehearsed take is the most honest read on where the contrast actually stands.