The acoustic features of speech exhibit variations across different respiratory conditions, highlighting the potential of voice analysis as a valuable tool for non-invasive monitoring systems. Early detection of exacerbations plays a critical role in the effective management of chronic respiratory diseases, such as chronic obstructive pulmonary disease (COPD) and asthma. This paper presents the utilization of fused acoustic features from multiple domains, integrated with a Locally-enhanced Feed-Forward Network (LeFF) Transformer model, to classify exacerbated and stable speech in COPD and asthma patients. The proposed methodology is evaluated on the TACTICAS dataset, demonstrating superior performance compared to current state-of-the-art approaches, underscoring its potential for exacerbations monitoring in COPD and asthma patients.