Abstract

This paper presents a novel speech recognition system specifically designed for processing structured numerical sequences in Vietnamese language within grid-based coordinate systems. The proposed system introduces three key innovations: a specialized recognition framework for grid-based coordinate structures, an adaptive noise filtering mechanism for robust performance in noisy environments, and optimized real-time processing capabilities. Using a 5x5 grid system as a demonstration platform, our implementation achieves 75.8% accuracy for complete coordinate sequences in office environments (SNR ≈ 25dB) and maintains usable accuracy of 70.5% in high-noise conditions (SNR ≈ 12dB), with average processing latency under 450ms. The system demonstrates practical viability for real-world applications requiring structured numerical sequence recognition in challenging acoustic environments.