Functional near-infrared spectroscopy (fNIRS) is now a common tool in neuroscience research, but due to the different analytical methods available, it is challenging to determine the best approach for data analysis. We investigated the effect of short channel (SC) correction on a block design experiment with auditory, visual, and audio-visual stimuli, by removing non-neuronal signals from fNIRS long channels. We tested three analysis pipelines with and without SC correction. fNIRS signal was deconvoluted using Generalized Linear Model (GLM) in which orthogonalized SC signal (first analysis), nearest SC signal (second analysis), and no SC signal (third analysis), was added as regressor of non-interest. Sixteen participants (from 24 to 34 years old) were recruited. All tasks were passive. There were three conditions: i) visual (V), ii) auditory (A) and iii) audio-visual (AV). We used dynamic (gif) ecological stimuli such as moving car or a ringing phone. Stimuli were presented for a duration of 10 seconds in successive blocks containing each of the three conditions. A silent jitter interval of 5 to 10 seconds was added between each stimulus. Optodes were located bilaterally on the temporal (over the auditory cortex) and occipital (visual) areas. Results showed significant differences between betas when comparing A, V and AV conditions from the two analysis with SC, the analysis with orthogonalized SC provided the best result. Data analyzed without SC did not provide usable results. This study highlights the critical role of SC in fNIRS in removing systemic signal, with orthogonalized SC proving most effective.