UUS™ · Ring 1 · Pronunciation

· 40+ languages

Hear the gap.
Then close it.

UUS™ — Ultra-unit Scoring — measures every phoneme in every utterance against your dialect model, then shows you exactly which sounds are costing your ring. Not pass/fail. A precision deviation map.

"You cannot acquire a sound you have never consciously noticed deviating." — Schmidt (1990), Noticing Hypothesis

UUS™ · Phoneme Analysis
73/100

Target phrase

Je voudrais un café

Native
You

Phoneme breakdown · tap to inspect

Four layers in every utterance

01

Phoneme deviation

Each phoneme scored against your target dialect. Québécois /ʀ/ is not Parisian /ʀ/. UUS scores against the specific dialect you train.

02

Prosody & intonation

Azure Speech captures intonation, stress, and connected speech — not just segmental accuracy. Your vowel might be right; your sentence melody may still mark you as foreign.

03

Waveform comparison

Your utterance overlaid against the native waveform. You see where the shapes diverge before you drill — not an abstract score.

04

HEXI prioritisation

HEXI reads your UUS deviation map every session. The phoneme furthest below threshold enters your Imprint™ queue first.

Four workouts · one ring

UUS™ is the scoring engine. These are the four modes that use it.

01 · Hear → Repeat → Score

Imprint™

→

One phoneme. One phrase. Scored at the phoneme, word, and sentence level against your dialect model. Not pass/fail — a deviation map. HEXI routes the weakest phoneme to the next session automatically.

"The gap isn't your accent. It's three sounds you've been rounding off."

02 · Target → Drill → Compare

Phoneme Focus

→

Your weakest UUS phonemes, isolated. The drill re-uses your real recorded phrases as comparison material — not a generic model speaker. HEXI picks the target phoneme based on your ring history.

"Drill the sound you're missing. Not the one that feels hard."

03 · Listen → Mirror → Score

Shadow & Echo

→

Hear the native phrase once. No transcript. Echo it back. UUS scores your echo against the dialect model — intonation, stress, connected speech. Text suppressed so your ear leads, not your eye.

"Pronunciation has a body. Train yours to listen first."

04 · Communicate → Pronounce → Score

Scenario Voice

→

A real communicative pressure situation. Pronunciation scored inside the flow of meaning — not in sterile isolation. The café doesn't pause while you think about your /ʀ/.

"The café doesn't wait for you to feel ready."

Under the hood

Dialect-specific,
not generic.

UUS™ is powered by Azure Speech's phoneme-level pronunciation assessment — the same technology used in clinical speech therapy. Voicely routes it through dialect-specific acoustic models so your Québécois /ʀ/ is scored against Québécois speech, not generic French.

Azure Speech

Phoneme, word, and sentence granularity — plus prosody and intonation on every utterance.

Dialect models

Québécois ≠ Parisian. Colombian ≠ Castilian. Scored against the exact dialect you train.

UUS Score

Not a pass rate — a deviation map. Each phoneme gets a 0–100 precision score.

HEXI routing

Weakest phonemes surface first every session. The ring fills as deviation shrinks.

The research

Schmidt (1990)

Noticing Hypothesis

Acquisition requires conscious noticing of the gap between your output and the target form. UUS makes the phoneme deviation visible — the noticing is built into the feedback loop.

Flege (1995)

Speech Learning Model

Adults form new phoneme categories through perceptually distinct, repeated exposure to the target. Dialect specificity in scoring is the mechanism — not a feature. Québécois /ʀ/ requires a Québécois model.

Munro & Derwing (2001)

Intelligibility vs Accentedness

The goal of pronunciation training is communicative intelligibility — not accent erasure. UUS scores the phonemes that affect whether you are understood, not those that simply mark you as foreign.

Hear the gap.

Then close it.

Score your pronunciation →