In detecting replay spoofing attacks against voice-controlled systems such as smart speakers, the utilization of multi-channel information is crucial. ReMASC is the only large-scale multi-channel replay attack corpus created with multiple recording devices featuring different microphone array configurations, enabling comparison of detection performance across different device configurations. However, a fair comparison is not possible because the composition ratios of recording conditions are inconsistent across recording devices. Additionally, since conditions other than speakers are known in the existing subsets, the independent analysis of each recording condition’s impact on performance is not feasible. In this study, we redesign ReMASC by introducing data cleaning to unify the composition ratios of recording conditions across recording devices and a subset splitting method that allows arbitrary recording conditions to be controlled as known or unknown. Furthermore, we conducted experiments based on the redesigned data splits and demonstrated that it is possible to quantitatively evaluate the impact of individual recording conditions and device configuration differences on detection accuracy.