This paper introduces a novel method, VOLEMIA, to improve blood pressure (BP) estimation from the photoplethysmography (PPG) signal. Existing literature has often relied on long-duration PPG signals, which can be noise-prone, thereby compromising the performance of BP estimation. As a solution, VOLEMIA presents the PulseBlend Deconstructor (PBD), which partitions the lengthy PPG signal into shorter segments and consolidates the segments to extract the noise-resilient PPG signal. Furthermore, VOLEMIA presents the Pulse Spectra Extractor (PSA) mechanism to extract pulsatile spectral features from the PPG signal because spectral features provide relevant cues for systolic BP (SBP) and diastolic BP (DBP). Unlike existing methods, VOLEMIA incorporates these features into an advanced sequential deep learning framework while also considering the correlation between SBP and DBP. A new composite loss function is proposed to enable the network to learn both individual and correlated BP features, enhancing performance. Experimental results on our newly designed DILPPG and publicly available MIMIC-II dataset demonstrate that VOLEMIA exhibits superior performance than the existing methods across both datasets. Also, it indicates that key components of VOLEMIA, like PBD, PSA, and composite loss function, play a crucial role in performance improvement. Dataset link: https://github.com/TrishnaSaikia/-DILPPG-Dataset.git