Voice biometrics have been commonly employed in different application domains for user verification. The rise of voice spoofing attacks has challenged the current automatic speaker verification (ASV) technology. Voice replay or presentation attacks are the easiest to generate and launch on any voice-operated application to get any system’s access by failing the ASV system. Existing antispoofing solutions focus on accurate detection of spoofing attacks without considering the computational cost which makes them inappropriate for resource-limited devices, thus, bringing a need to develop more efficient antispoofing solutions. This paper presents an efficient voice replay antispoofing method capable of reliable real vs replay audio classification. We proposed an efficient ten-dimensional fused feature set obtained, after applying a two-stage feature selection method, from the 39-dimensional spectral features of 13 different categories. The employment of Pearson correlation coefficient and variance inflation factor makes our feature selection method to identify the ten most effective features from the GTCC, MFCC, and other spectral features. This fused feature set is used to train the SVM to classify real/bonafide or replay samples. The competency of our method was tested on two benchmark ASVspoof2017-PA and ASVspoof2019-PA datasets. Experimental results reveal that our method surpasses the performance of baseline antispoofing countermeasures. The outcomes of a detailed experimental study demonstrate the efficacy of our replay antispoofing method that can be deployed in resource-limited devices.