Enhanced Arabic Human-Machine Dialogue Using a Two-Level Dynamic Programming Algorithm

This paper presents a prototype man–machine dialogue system specifically designed for Arabic, addressing the growing need for voice-based interaction in under-resourced linguistic contexts. Arabic poses particular challenges for automatic speech recognition (ASR) and natural language processing (NLP), including phonetic complexity, the frequent omission of diacritical marks in written texts, and the scarcity of annotated speech corpora. These factors have significantly impeded the development of robust Arabic voice interfaces. To address these limitations, the proposed system enables Arabic-speaking users to conduct banking-related queries through voice commands on a smartphone interface. The system incorporates two complementary feature extraction techniques—Mel Frequency Cepstral Coefficients (MFCC) and Perceptual Linear Prediction (PLP)—and employs a two-level dynamic programming algorithm to iteratively align acoustic feature vectors using Euclidean distance. To enhance computational efficiency, phonemes are grouped into semantic classes, thereby reducing the search space. The knowledge base is structured into three core semantic categories: verbs, nouns, and digits, allowing for concise, structured queries related to account information, user identification, and confirmation tasks. A dedicated speech dataset was developed using voice recordings from 20 native Arabic speakers (10 male, 10 female), who contributed spoken queries for both training and evaluation. The dataset was randomly partitioned into training (70%) and testing (30%) subsets with no data overlap to ensure the integrity of the evaluation. Experimental results show a sentence comprehension accuracy of 92.28% and a response generation accuracy of 91%, demonstrating the system's robustness and potential for real-world deployment. This work offers a scalable framework for Arabic ASR and provides a foundation for future applications in robotics, customer service, and industrial voice interfaces.