// THESIS GUIDANCE PORTAL · Computer Science

NLP-Based Nepali Sign Language Translator

Level: Bachelor, MasterDifficulty: Advanced★ Popular Choice

1. Introduction & Problem Statement

Overview: Computer-vision system translating Nepali Sign Language gestures into Devanagari text and speech for Nepal's deaf community.

Background Context (Nepal): Nepal's hearing-impaired population faces significant communication barriers in public offices, schools, and hospitals. Real-time vision-based sign translation builds an accessible digital communication bridge.

2. Research Objectives

›Collect and curate video gesture dataset of standard Nepali Sign Language (NSL) alphabet and frequent words
›Train MediaPipe hand landmark extractor combined with LSTM / Transformer gesture classifier
›Develop real-time webcam inference engine outputting Devanagari text
›Integrate text-to-speech engine generating spoken Nepali audio
›Evaluate translation accuracy, latency, and lighting robustness

3. Proposed Methodology

Dataset creation collaborating with local deaf welfare associations and special education schools
3D hand-keypoint extraction using Google MediaPipe Hand Tracking
Sequence classification using Bi-LSTM / Spatial-Temporal Graph Convolutional Networks (ST-GCN)
Devanagari NLP text post-processing and gTTS/eSpeak speech synthesis integration
Performance evaluation measuring top-1 accuracy, frame rate (FPS), and user satisfaction score

$ Worked Example / Sample Scenario

Sample Scenario: A deaf citizen presents sign language gestures in front of a laptop webcam at a ward office. The MediaPipe model tracks 21 hand keypoints, converts gestures into Devanagari 'नागरिकता प्रमाण-पत्र' (Citizenship Certificate), and reads it aloud via speaker.

4. Thesis Chapter-by-Chapter Outline

Chapter 1: IntroductionTU/KU standard

Background, problem statement, research questions, objectives, scope, limitations, and significance of the study

Chapter 2: Literature ReviewTU/KU standard

Theoretical framework, conceptual models, previous empirical studies in Nepal and developing nations, MediaPipe hand keypoint tracking and recurrent neural network gesture models, and gap analysis

Chapter 3: Research MethodologyTU/KU standard

Research design, population/sampling framework, data collection instruments, analytical tools, and ethical considerations

Chapter 4: Data Analysis & ResultsTU/KU standard

Empirical findings, statistical testing, model estimations, gesture confusion matrix and frame-rate benchmarks, and detailed discussion

Chapter 5: Conclusion & RecommendationsTU/KU standard

Summary of key findings, theoretical contributions, policy recommendations, and future research directions

5. Recommended Tools & Technologies

To implement the practical, technical, or analytical portions of this thesis topic, the following software tools, libraries, or APIs are recommended:

PythonOpenCVMediaPipePyTorch / TensorFlowStreamlit / ReactgTTS

6. Core References & Academic Sources

[1]National Association of the Deaf Nepal (NADN) — Nepali Sign Language Dictionary
[2]IEEE Xplore / Google Scholar — Sign language translation using MediaPipe and LSTMs
[3]NepJOL — Computer vision research in Nepali script and gesture recognition

7. Frequently Asked Questions (FAQs)

Q: How does Nepali Sign Language (NSL) differ from American Sign Language (ASL)?

NSL has unique regional gestures, hand shapes, and grammar rules tailored to Devanagari structure that differ significantly from ASL.

Q: Do I need a high-end GPU to run hand keypoint detection?

MediaPipe Hand Tracking is optimized to run at 30+ FPS on standard CPU laptops without requiring a discrete GPU.

Q: Where can I find NSL dictionary video references?

The National Association of the Deaf Nepal (NADN) publishes official NSL dictionaries and instructional material.

Q: Is static gesture recognition enough for a Bachelor thesis?

Recognizing static Devanagari finger-spelling (alphabet) is suitable for a Bachelor thesis, while dynamic word sequence translation fits Master level.