Noise reduction is a fundamental aspect of stress electrocardiogram $(ECG)$ recording. In this setting, muscular noise represents the main antagonist to signal quality. A possible solution to muscle noise in stress $ECG$ is to exploit the information redundancy in 12 - lead recordings to reduce noise while preserving the $ECG$ signal. Source Consistency Filtering $(SCF)$ is a spatial redundancy filter that follows this principle. In this paper, we compare the muscle noise rejection performance of conventional $25Hz$ and $40Hz$ low-pass filters (LPFs), the SC $F$ ‘ and a method based on singular value decomposition $(SVD)$ which exploits both the spatial and temporal correlation in the $ECG$ signal. Our results indicate that the $SCF$ can afford a $QRS$ complex distortion lower than that of a 40 $Hz$ lowpass filter while still maintaining a high noise rejection. The $QRS$ detection accuracy on the filtered $ECG$ was comparable for all methods except for the $SVD$ filter, which allowed a superior detection performance score in all the records.