⌂ Home
Collocation Pronunciation

Master the rhythm of
English collocations

Hear the phrase spoken by Google US English — then imitate it. Your score is based on how closely your pronunciation matches the model voice. Instant results, no AI judge needed: Google's own speech engine decides.

TTS gold standard Instant scoring Google US English model Hear it first, then speak 4 scoring dimensions 75+ collocations · 3 levels
Practice
Requires Chrome or Edge for microphone
—
Average
0
Done
—
Best
0
Streak 🔥
verb + noun
—
Speech recognition unavailable — use Chrome or Edge.
Allow microphone access, then press Speak
You said:
—
—
🇺🇸 Google Match 45%· 🎯 Word Accuracy 25%· ⏱ Speaking Pace 20%· ᴸᴿ L/R Distinction 10%
🇺🇸
Closeness to Google US English
—
🎯
Word Accuracy
—
⏱
Speaking Pace
—
ᴸᴿ
L / R Distinction
—
How it works
01
Hear the model
A collocation appears and Google US English automatically speaks it at a clear, natural pace. The Speak button is locked until the model finishes — you must listen first. Press Play again to hear it as many times as you need.
02
Imitate it
Press the microphone button and say the phrase, trying to sound exactly like the model voice. Focus on matching the stress pattern, vowel sounds, and rhythm — not just the individual words.
03
See your match score
Results appear instantly across 4 dimensions: Google acoustic match (45%), word accuracy (25%), speaking pace (20%), and L/R distinction (10%).
Scoring components
🇺🇸
Closeness to Google US English 45% — PRIMARY
The most important dimension. Chrome's speech recognition engine uses the same acoustic model as the Google US English TTS voice. When you speak and the engine is highly confident it understood you, it means your speech closely resembles the model voice. A score above 80 means you sound very close to native American English. Speaking word-by-word with equal stress on every syllable typically scores below 50 — the phrase must flow as one natural unit, just like the model.
🎯
Word Accuracy 25%
Checks whether the speech engine recognised the correct words from your attempt. Up to 5 alternative interpretations are requested and the best-matching one is used for scoring. Each unrecognised word reduces this score proportionally. If several words are misrecognised, it indicates pronunciation differences that confused the engine — focus on those words when you listen to the model again.
⏱
Speaking Pace 20%
Compares the duration of your speech to the Google TTS model, which speaks at a natural connected pace (rate 0.88). Speaking too slowly — word by word with pauses between them — reduces this score, as does speaking too quickly and slurring words together. Aim to finish within ±40% of the model's duration. Listening to the model several times before speaking will help you internalise the natural rhythm and speed.
ᴸᴿ
L / R Distinction 10%
Japanese has a single flap sound (ɾ) that maps to both English /l/ and /r/, causing frequent confusion. This dimension checks whether a word containing /l/ or /r/ was recognised as its L↔R swapped counterpart. Examples: "right" heard as "light", "road" heard as "load", "play" heard as "pray". /l/ is lateral — the tongue tip touches the alveolar ridge. /r/ is retroflex — no tongue-tip contact, lips slightly rounded. Scores N/A if the phrase contains no L or R sounds.

Why collocations & why rhythm?

A collocation is a pair or group of words that habitually occur together — like "make a decision", "heavy traffic", or "totally committed". Native speakers acquire these as chunks. Learners who speak each word separately with equal stress sound immediately non-native regardless of how correct individual sounds are.

Rhythm in English is stress-timed: content words carry full vowel quality while function words reduce. Mastering this alternation is the single biggest leap toward natural-sounding English — which is why rhythm still carries 25% of your score.

The L/R distinction check targets the most reliably detectable Japanese pronunciation error: confusing /l/ and /r/ produces a different real word (e.g. "road"→"load") that the speech engine records in its transcript. Practise minimal pairs — right/light, road/load, red/led, rice/lice — until the difference feels automatic.

Practice tips
Say the whole collocation as one breath group — it is the unit, not individual words.
Use the rhythm pattern (● ○ ○●○) as a drumbeat guide before speaking.
For clusters: practise br-, cl-, str-, pr- in isolation first. Keep the consonants tight — no vowel between them.
Reduce function words aggressively — "a" is always /ə/, never /eɪ/ in connected speech.
For American accent: let your r's colour the surrounding vowel and tap the T between vowels.
Speech recognition requires Chrome or Edge. Grant microphone permission when prompted.