JYUTONE Tone-aware
Six contours, including checked tones
Pronunciation coaching for Cantonese
Speak Cantonese,
with Jyutone.
Get precise feedback while you speak—not another lesson to sit through. Jyutone maps every syllable against Cantonese’s six tone contours, including the checked tones traditionally counted as seven to nine. Practise for 10–30 minutes a day and make accurate pronunciation feel automatic.
- Purpose-built speech model
- Focused drills
- Progress stays on-device
Cantonese speech intelligence
Four-part scoring
Sound · tone · completeness · fluency
Frame-level alignment
Locate the exact sound that drifted
Inside the engine
From a waveform to one useful next step.
We do not hide the result behind a single score. Each attempt is cleaned, aligned to the target, measured across separate dimensions, and turned into the next drill your speech actually needs.
01Capture
Start with a cleaner signal.
Speech is separated from silence and background energy, so the model can focus on the part that matters.
waveform → speech window02Align
Place every sound on the timeline.
The recording is matched with the reference at sound level, locating initials, finals, and syllable boundaries in time.
audio × reference → timeline03Score
Measure sound and tone separately.
Pronunciation, pitch contour, completeness, and fluency remain distinct, so one strength cannot conceal another weakness.
4 signals → precise feedback04Adapt
Turn each slip into the next drill.
Sounds that need work are added to your on-device practice history, so the next session focuses where it can make the most difference.
attempts → personal practiceBuilt around Cantonese
What general-purpose speaking apps miss.
Most speaking apps are not tuned to the sound system of Cantonese. Jyutone uses a purpose-built speech model to locate the syllable, sound, or tone that needs work, then turns it into a focused drill. Your practice history stays on your device; scoring audio is processed only when requested and is not kept as a long-term voice archive.
01 · Our model
A speech model built for Cantonese.
This is not translation or generic speech-to-text. The model is trained on Cantonese speech, aligns your recording with the reference phrase, and pinpoints the initial, final, or tone that drifted.
- Sound-level alignment, not just an overall score
- Separate scores for pronunciation, tone, completeness, and fluency
- Pitch tracking across all six tone contours
Sound alignment · 唔該晒
02 · Speaking gym
Work on what needs work—not the whole lesson again.
Initials, finals, and tones are tracked separately. If ng- keeps disappearing or tone 4 never settles in the low register, jump straight into a drill for that sound instead of repeating an entire chapter.
- Sounds grouped by where they are formed
- Jump from a weak sound straight into useful phrases
- A daily set alongside focused pronunciation drills
Weak spots · drill today
03 · On-device privacy
Your practice history stays on your device.
When you request a score, that clip is sent for processing and is not kept as a long-term voice archive. Check-ins, weak sounds, and practice history stay on your phone. Jyutone does not build a cloud voiceprint or use your speech to train third-party models.
- Progress and pronunciation records stay on-device
- No cloud account required to practice
- Scoring audio is processed, not archived
Where the data lives
04 · Practice log
See the progress in every attempt.
Each character and sound is tracked over time. Instead of seeing only “82 today”, you can watch 「晒」 move from 62 to 82 and check whether tone 4 stayed steady this week. Visible progress makes it easier to keep going.
- Results for every character and sound
- A practice calendar that makes consistency visible
- Weak-sound trends at a glance
「晒」saai3 · last three tries
Core · Correction
Know exactly what to fix.
Repeat a Cantonese phrase and the model scores every syllable for pronunciation, tone, completeness, and fluency. It then points to the character and tone that need another try—details that are difficult to judge by ear on your own.
- A purpose-built model aligned to Cantonese sounds
- Pitch traces across six contours
- Pinpoints the character and the tone
Six contours · pitch reference
詩si1 high level
史si2 high rising
試si3 mid level
時si4 low falling
市si5 low rising
是si6 low level
Core · Speaking practice
Build speaking habits in 10–30 minutes a day.
Start speaking straight away with a daily set and focused drills: listen → repeat → record → review. As accurate sounds and tones become familiar across everyday vocabulary, speaking starts to feel far less daunting.
- Short, frequent sessions that fit a workday
- Today’s set + drills for the sounds you miss
- Learn through real phrases, with pronunciation first
How one session works
1Hear the demo
2Say it aloud
3Record it
4Read the correction
New · Scripts
Write what you mean. Practice Hong Kong Cantonese.
Type what you want to say in Mandarin or English. Jyutone turns it into natural Hong Kong Cantonese and provides a reference reading, so you can practise—and receive feedback on—the phrase you actually need.
- Mandarin / English → Hong Kong Cantonese
- A reference reading for every phrase
- Save lines and loop a playlist
Rewrite preview
Core · Feedback
Every recording has a next step you can take.
The score is only the start. You will see which tone fell too low, where a syllable ran long, and what to try next. Every attempt updates your on-device history, and sounds that need work return in a later drill.
- Tone contrast at a glance
- Clear, actionable coaching
- A result for every sound
Sample feedback
Total 82 · tone weakens at the end
✓唔 / 該 — pronunciation steady
!晒 saai3 — 3rd tone is low; don’t drag it
Why Jyutone
Versus a typical course or a generic app
The difference is not more course material. It is immediate, useful speaking feedback and a practice routine designed around trying, adjusting, and trying again.
Typical course / generic app
- Long lessons before you get to speaking
- A general speech model not tuned for Cantonese tones
- The course moves on while weak sounds stay unresolved
- One overall score with little guidance on what to fix
Jyutone
- Start with real sounds and tones—speaking comes first
- A Cantonese-first speech model with sound-level feedback
- Focused drills for the sounds that need more work
- Turn your own ideas into Hong Kong Cantonese you can practise
- Practice history stays on your phone; scoring audio is not archived
Open learning notes
See the sound, then hear it differently.
Three interactive, plain-English guides connect Jyutping on the page, the acoustics behind a recording, and the signal processing that prepares speech for scoring.
Start speaking—and make every tone count.
On iPhone, continue to the App Store. On Android, download the app directly. Less theory, more useful speaking practice.