Design and evaluate a mobile app for real-time dialect identification

Thesis Abstract

Language variation across regions poses significant challenges and opportunities for effective communication, linguistic documentation, and cultural preservation. Despite the proliferation of digital communication tools, there remains a notable gap in accessible, real-time dialect identification technologies that can support linguists, language learners, and speech recognition systems. This study aims to design, develop, and evaluate a mobile application capable of accurately identifying dialects in real-time using speech input. The specific objectives include analyzing phonetic and acoustic features distinctive to targeted dialects, implementing machine learning algorithms for dialect classification, and assessing the application's accuracy and usability in diverse contexts. The research adopts a mixed-methods approach, combining quantitative and qualitative methodologies. A descriptive research design guides the development and evaluation stages, emphasizing iterative testing to refine the application's algorithms. The population encompasses native speakers from three distinct dialect regions, totaling 300 individuals, with a sample size of 150 participants selected via stratified random sampling to ensure representativeness across dialectal variations. Data collection instruments comprise recorded speech samples, which are processed through spectrographic analysis and phonetic transcription to identify salient features. These features serve as input variables for supervised machine learning models, including support vector machines (SVM) and convolutional neural networks (CNN), trained and validated using a 70/30 split, with cross-validation to prevent overfitting. Data analysis involves both the examination of model performance metrics—such as accuracy, precision, recall, and F1-score—and thematic analysis of user feedback obtained through semi-structured interviews and usability questionnaires. The quantitative assessment determines the effectiveness of the classification algorithms, while qualitative insights inform interface design improvements and user satisfaction levels. The study hypothesizes that machine learning models, when trained on carefully extracted phonetic features, can achieve a minimum of 85% accuracy in dialect identification, with CNN outperforming SVM in classification tasks. The research applies the Variability Hypothesis in phonetics and the Theory of Speech Variation to underpin feature selection and model development. Expected findings suggest that the mobile app can reliably differentiate dialects with high accuracy, providing rapid, on-the-fly feedback to users. The study anticipates that advanced deep learning models will significantly improve classification precision, offering a viable tool for dialectologists, language educators, and technologists. Furthermore, usability evaluations are expected to reveal critical design factors influencing user engagement and acceptance. This research contributes novel insights into integrating phonetic feature analysis with contemporary machine learning techniques for dialect identification in mobile applications. It advances theoretical understanding by empirically validating the applicability of speech variation theories in digital language technology and establishes practical frameworks for deploying real-time dialect classifiers in resource-constrained environments. The findings will inform future research on scalable dialect recognition systems, promoting linguistic diversity and technological inclusivity. The study concludes by recommending the integration of the app into educational platforms, broadcast media, and linguistic research tools, while emphasizing the importance of ongoing algorithm refinement through expanded datasets. Limitations include potential biases inherent in speech recordings and the challenges of dialectal overlap. Future work should explore multilingual capabilities, dialectal continua, and the application of unsupervised learning models to accommodate under-resourced dialects. Overall, the research demonstrates that a carefully designed mobile application can significantly enhance the accessibility and accuracy of dialect identification, fostering a deeper understanding of linguistic variation through innovative technological solutions.