Dot-Trojan-Horse/NLP-Research-Project

Error Analysis of isiZulu Street Name Pronunciation in Existing GPS Systems. The main focus is to evaluate the performance of current systems in handling isiZulu pronunciation, later develop and train speech models using the collected dataset in order to improve the pronunciation accuracy of isiZulu street names in South African navigation system # NLP Research Project ## Apple Maps Pronunciation Evaluation for isiZulu Street Names This repository contains the implementation developed for my BSc Honours Computer Science research project the pronunciation of isiZulu street names in Apple Maps using objective speech processing techniques and qualitative linguistic error analysis. ## Project Title Error Analysis of isiZulu Street Name Pronunciation Using Speech Processing Techniques ## Overview This project evaluates the pronunciation quality of Apple Maps Text-to-Speech (TTS) against native isiZulu speaker recordings. The evaluation combines objective speech processing techniques with manual linguistic analysis to identify pronunciation errors affecting isiZulu street names. ## Methods The project includes: - Dynamic Time Warping (DTW) - Mel-Frequency Cepstral Coefficients (MFCC) - Pronunciation Accuracy Score (PAS) - Pitch Analysis - Duration Analysis - Energy Analysis - Spectrogram Analysis - Mel Spectrogram Analysis - MFCC Heatmap Comparison - Manual Error Taxonomy ## Repository Structure ```text data/ figures/ modules/ output/ reports/ results/ scripts/ ``` ## Software Python 3.12 Main libraries: - librosa - numpy - pandas - matplotlib - scipy - openpyxl ## Ethical Statement Ground truth recordings from Native speakers are not included in this repository to protect participant privacy and comply with research ethics. Sample data or recordings can be provided upon reasonable academic request, subject to ethical considerations. ## Author Sam Khumalo BSc Honours Computer Science University of South Africa