Welcome To The Voca Guide

There is a clear research gap in the development of AI-powered spoken language-learning applications for visually impaired children in Sri Lanka. While global solutions in speech recognition and grammar correction have advanced rapidly, very few are tailored to the unique linguistic, cultural, and accessibility needs of Sri Lankan learners. Current approaches often fail to capture the phoneme-level detail and grammar variation present in Sri Lankan English, and existing gamification methods remain heavily visual, leaving blind learners at a disadvantage. To bridge these gaps, dedicated solutions like VocaGuide are essential, ensuring that accessibility is not treated as an afterthought but as the foundation of design.

1. Audio-Only Gamification for Visually Impaired Learners

* Many language-learning apps focus on visual gamification, making them unsuitable for visually impaired children.

* Current tools rely on static question banks and generic feedback, with little adaptation to individual learner errors.

* There is a lack of real-time, speech-driven micro-quizzes and spaced repetition delivered through natural, child-friendly TTS.

2. Mispronunciation & Grammar Error Detection

* Existing systems report sentence-level accuracy but lack phoneme-level diagnostics that highlight segmental and suprasegmental issues common in Sri Lankan English.

* Current classroom products rely on visual feedback (charts, heatmaps) rather than concise audio explanations accessible to blind learners.

* Grammar correction tools tend to over-correct Sri Lankan English forms, lowering trust and usability.

3. Conversational Grammar Error Correction

* Most grammar correction systems ignore conversational context, cultural nuance, and pragmatics, leading to rigid or irrelevant feedback.

* Existing models struggle with efficiency on mobile devices, often suffering from latency and high memory demands.

* Accessibility gaps remain, as few systems are designed to be speech-first, low-latency, and inclusive of multiple disabilities.

The main research problem addressed in this study is the lack of accessible, audio-first language learning tools for visually impaired children in Sri Lanka. Existing mobile applications for language learning are predominantly visual, making them unsuitable for children who are blind or have low vision. As a result, these learners have very limited opportunities to practice spoken English in an engaging and accessible environment.

Current speech recognition systems often fail to capture the unique features of Sri Lankan English pronunciation, overlooking common segmental and suprasegmental errors that affect communication. At the same time, generic grammar correction models frequently misidentify valid Sri Lankan English usage as incorrect, which can discourage learners and undermine trust in the system. Most available solutions also depend on reliable internet connectivity and high-end devices, which makes them inaccessible in low-resource educational settings. Together, these issues leave visually impaired learners without credible, personalized support for improving their spoken language abilities.

Our research aims to design and evaluate an inclusive, speech-first tutoring system that provides accent-sensitive pronunciation feedback, delivers trustworthy grammar suggestions adapted to Sri Lankan English, and incorporates audio-based gamification to keep learners motivated. The solution will operate effectively on mid-range mobile devices, function offline when needed, and prioritize user privacy. By addressing these challenges, the project seeks to create a practical and scalable tool that improves language skills while ensuring accessibility for visually impaired children in Sri Lanka.

Main Objective

The main objective of this research is to design, develop, and evaluate an audio-only, conversational, and gamified English language learning system tailored for visually impaired Sri Lankan learners. The system aims to provide accent-aware pronunciation support, context-sensitive grammar correction, and interactive quiz-based practice, all delivered through fully accessible, speech-first interfaces. By focusing on accessibility, cultural relevance, and privacy, the solution seeks to empower learners to practice independently, build confidence in spoken English, and achieve measurable educational outcomes under real-world resource constraints such as limited connectivity and mid-range devices.

Sub Objectives

1. Design an adaptive quiz-based gamification engine that generates voice-only practice activities from detected weaknesses, with spaced repetition for sustained progress.

2. Implement a conversational AI assistant that supports multi-turn spoken interaction, role-play scenarios, and peer-style dialogue to encourage natural language use.

3. Develop a phoneme-level pronunciation assessment module that detects and explains mispronunciations in Sri Lankan English with accessible audio feedback.

The methodology for developing Vocaguide involves a mixed-methods approach. First, we will conduct qualitative research through interviews and focus groups with visually impaired learners and educators to identify challenges in spoken language learning. Second, we will carry out a quantitative phase using surveys and pilot studies to collect data on learner needs, technology use, and baseline performance.

We will then train and fine-tune speech and language models for phoneme recognition, pronunciation scoring, and grammar correction, and implement these in a React Native + FastAPI application designed with accessibility-first principles. Usability testing and iterative prototyping will be conducted to refine the system. Finally, learning outcomes will be evaluated through pre- and post-test comparisons, usage analytics, and learner feedback to ensure effectiveness and continuous improvement.

React Native (Expo)

FastAPI

Python

Jupyter Notebook

Anaconda Navigator

Postman API Platform

VS Code

Figma UI/UX

GitHub

ngrok