This document proposes an Enhanced Modulation Spectral Subtraction (EMSS) method for enhancing noisy speech signals to improve speech recognition performance in Internet of Vehicle Things (IOVT) systems. The EMSS method uses a two-stage analysis-modification-synthesis framework, applying short-time Fourier transform first at the acoustic level and then again at the modulation level to extract modulation spectra. At the modulation level, noise is estimated and subtracted from the noisy modulation spectra. Evaluation on speech emotion recognition tasks with different noise types shows the EMSS method improves the signal-to-noise ratio segmentation by 39-60% compared to traditional spectral subtraction and modulation spectral subtraction methods, achieving about a 50% better recognition performance.