Severe acute respiratory infections (SARI) remain a major global health concern, particularly in resource-limited settings where comprehensive pathogen detection is challenging. Conventional diagnostics, including culture and multiplex PCR, are restricted to predefined targets and may miss emerging or uncommon pathogens. This study aimed to develop and optimize a practical long-read metagenomic next-generation sequencing (mNGS) workflow using Oxford Nanopore Technology for sputum samples from adults hospitalized with SARI. We evaluated sputum liquefaction, host nucleic acid depletion strategies, and SMART-9N-based cDNA amplification to enhance microbial and viral nucleic acid recovery. We found that dithiothreitol (DTT) treatment significantly improved nucleic acid extraction. DNase I treatment effectively reduced host background while preserving both viral and bacterial sequences, outperforming filtration-based approaches that reduced viral recovery. The optimized workflow enabled unbiased detection and strain-level resolution of respiratory pathogens, including rhinovirus C1/C42, human coronavirus HKU1, Mycoplasma pneumoniae, Haemophilus parainfluenzae , and Pseudomonas aeruginosa , with high genome coverage. This approach demonstrated robust performance for respiratory pathogen identification directly from sputum samples. The proposed workflow is scalable and suitable for clinical diagnostics and public health surveillance. Further validation in larger cohorts is warranted to assess diagnostic sensitivity, accuracy, and feasibility for routine implementation.