PRODUCT

AI Oral Assessment

AI 口語評估 · Trilingual live speaking practice (Cantonese · Mandarin · English)

Trilingual AI oral coach natively trained for HK — Cantonese, Mandarin, and English. Built-in TSA / DSE / interview scenarios, recitation, picture-based storytelling, and real-time dialogue with SEN-friendly adaptations. At partner schools every student talks to AI Oral ~4 times a month; weaker students average +59% score gain.

TeachersStudents
CantoneseMandarinEnglishP3–P6S1–S6

Weaker students +59% · 4 AI convos/student/month · Trilingual HK Cantonese + Mandarin + English

TSA Oral Practice
Prompt

“Tell us about your favourite hobby.”

0:000:07
Overall
4.0/ 5
Clear enunciationConfident tone
Pronunciation
4.2/5
Fluency
3.8/5
Content
4.3/5
Vocabulary
3.9/5
Add more detail
Transcript

“Hi, my favourite hobby is playing basketball every Saturday with my classmates. It helps me stay healthy and happy.”

Weaker students +59%
4 AI conversations / student / month

Teacher pain · Essai solution

30-student class: teachers can’t give meaningful 1-on-1 speaking time to each kid.

AI as a personal practice partner. At partner schools, every student talks to AI Oral ~4 times a month.

General AI voice tools only handle Mandarin or English — Cantonese-native HK students have nowhere to practise.

Cantonese, Mandarin, and English trained natively. HK vocabulary, HK curriculum, HK pronunciations.

SEN and weaker students are shy to speak in front of peers and end up practising the least.

Private AI dialogue with SEN-friendly pacing, repeat prompts, and forgiving feedback. Weaker students averaged +59% score gain in in-school study.

Key features

Trilingual — Cantonese + Mandarin + English

Each language natively trained — not a Mandarin model with a Cantonese voice skin. HK vocabulary and pronunciation throughout.

TSA · DSE · interview scenarios

Built-in simulations: TSA read-aloud, DSE group discussion, secondary-school and university interview prompts.

Recitation practice

Dedicated mode for solo recitation — pronunciation, pacing, and tone scored per phrase, with demo playback.

Picture-based storytelling

AI shows a picture and guides the student to describe the scene and build a narrative, with content and expression feedback.

Real-time dialogue · SEN support

Actual back-and-forth conversation. SEN mode supports slower pace, repeat prompts, extended response windows, forgiving feedback.

School-based prompt bank

Built-in HK-curriculum oral prompts plus teacher-uploaded school-based questions and marking criteria.

How it works

1

Pick the scenario

Teacher assigns a prompt or students pick — TSA / DSE simulation, recitation, picture storytelling, or free dialogue.

2

Student talks with the AI

AI responds live in the chosen language. Pronunciation, fluency, content, and vocabulary scored in real time.

3

Instant feedback + class report

Student sees scored feedback immediately; teacher gets a class weakness heatmap and can export a Word report.

Language + grade coverage

 P3P4P5P6S1S2S3S4S5S6
Cantonese oral
Putonghua oral
English oral
+59%

Average score gain for weaker students after practice · 4 AI conversations/student/month at partner schools · Cantonese / Mandarin / English natively supported

HK primary Chinese teacher

“(1) The UI is clean and works equally well on desktop, tablet, and mobile. (2) Content draws heavily from TSA, which makes it great pre-exam practice for improving students’ oral skills.”

— HK primary school Chinese teacher

FAQ

Is the Cantonese actually HK-authentic?
Yes. The model is natively trained on HK Cantonese with mixed colloquial and book-style vocabulary (畢, 嘅, 你同佢, etc.). It is not a Mandarin model with a Cantonese voice skin.
What does SEN support include specifically?
Slower speech rate, keyword repeat prompts, extended response windows, forgiving feedback, and optional picture cues. Teachers can configure individualised SEN settings per student.
How is scoring broken down?
Per dimension: pronunciation, fluency, content, vocabulary, grammar (English) or dialogue logic (Chinese). Teachers can customise weights or upload the school’s own marking criteria.
Can we use our school’s own oral prompts?
Yes. Teachers can upload school-based prompts or type them in one-by-one; the AI adapts scoring and follow-up questions accordingly.
How was the "+59%" score gain measured?
Sampled from students who used Essai AI Oral for 8+ weeks, comparing in-school oral scores before and after. The weaker-ability group (originally scoring in the lower half) averaged +59% gain. Higher-ability groups saw smaller gains due to higher starting points.
Is student audio used to train AI models?
No. Audio is used only for that session’s scoring and feedback. Each school’s data is isolated and the contract forbids training on school data.
Can we fund this with the EDB HK$500k programme?
Yes. Essai meets EDB Circular 221/2025 procurement requirements. Book a strategy session to get the application template.

AI Tutor

General chat

How can I help?

Ask me anything or pick a suggestion below.