The recent rise of generative AI makes detecting audio deepfakes increasingly challenging. Deep learning techniques produce highly realistic fake audio that can deceive both humans and Automatic Speaker Verification (ASV) systems. This paper presents ParallelChain Lab's submissions to the ASVspoof 5 challenge, namely a voice anti-spoofing system and a spoofing-robust ASV system. We developed an ensemble architecture comprising models trained with various augmentation types, including waveform augmentations, mel-spectrogram augmentations, and vocoder synthesis. An extensive experimental evaluation confirms the efficacy of our systems, achieving minDCF of 0.2660 for the deepfake detection system and min a-DCF of 0.3173 for the spoofing-robust ASV system in the closed condition.