Xinjie Intelligent Control (Hangzhou)XINJIE INTELLIGENT CONTROL · HANGZHOU

HomeProject CasesBlind-Student AI Intercom Terminal

Blind-Student AI Intercom Terminal

Education & Research · AI Voice TerminalPrototype Verified

Blind-Student AI Intercom Terminal

A handheld AI voice terminal designed for special-education and blind schools. Six physical buttons with voice feedback cover every operation — no screen needed: hold the side key to talk with the AI, English read-along scored instantly.

01 / PROJECT OVERVIEW

Project Overview

01

Physical-Button Accessible UX

02

Custom ESP32-S3 Firmware

03

4G / WiFi Dual Network

04

Instant Read-Aloud Scoring

05

Dual-Partition OTA

For visually impaired students, the first barrier of a learning device is not the content but "how to operate it". Most devices today rely on touch screens — open an app, scroll a list, find the button. Screen readers can read text aloud, but in a classroom that demands fast, quiet, one-handed operation, locating buttons by swiping is too costly for daily use. Meanwhile, spoken-language training that requires repeated speaking and instant correction is hard to do one-on-one in a class of dozens: average speaking time per student is under a minute per lesson, and after class nobody corrects pronunciation, so mistakes get locked in. This project tackles exactly these two problems — bring control back to the hands, and feedback back to the ears.

03 / OUR ENGINEERING

Our Engineering

The system has three layers; the device only handles content fetching, playback, recording, uploading and display, while content and scoring run in the cloud. The terminal layer is custom ESP32-S3 firmware integrating voice capture and playback, GUI, key scanning, 4G/WiFi communication, Bluetooth audio, power and status indication. The service layer handles authentication, heartbeat and battery reporting, binding notifications, remote firmware upgrades and pronunciation-scoring task scheduling. The content layer manages textbooks, units, sentences and audio resources plus learning progress and scoring records. The device's design constraint is a single rule: every daily operation must be possible without looking at the screen; the display exists only for low-vision students and teachers. Interaction: six physical buttons with fixed key positions that never change per page — the highest-frequency "hold to talk" is a large textured key on the left. Feedback: every action gets an audible response; four Chinese voice prompts are compiled into the firmware, independent of network or resource packs. Display: pure black background, high contrast, large dot-matrix Chinese font.

05 / DELIVERY

Delivery Results

Delivery Status

Prototype Verified

Prototype verified · main flows tested on hardware

Conclusion

Several hard engineering wins: audio is downloaded and cached per sentence (~9.6 KB each) so the first sentence plays in ~1.3 s with no waiting between sentences; a 500 ms low-frequency lead-in tone before Bluetooth playback eliminated the long-standing "first 2-3 syllables cut off" problem; silent failures caused by memory exhaustion (keys unresponsive, prompts not playing) were fixed by moving large task stacks and buffers to external RAM, lifting usable internal memory from 4-16 KB to 63-72 KB with stable long-run tests. The four-state AI conversation UI, role switching, the full five-page English listening-speaking flow, scoring upload with async polling, auto-continuous playback, Bluetooth/wired headset switching with speaker muting, RGB status LED, 4G/WiFi dual-network and dual-partition OTA all run on hardware.

Engineering first, quote later

Need a similar project?

Describe the scenario and delivery requirements — an engineer will assess feasibility and timeline directly.

ContactPhone