<p>Velocity-dependent dampers play a critical role in the seismic design of important structures by dissipating energy and mitigating structural vibrations during strong earthquakes, thereby reducing displacement and drift demand to protect structures in high-seismic-risk zones. An accurate estimation of the Spectral Velocity (SV) spectrum is vital for optimizing damper performance and ensuring structural resilience under dynamic seismic loads. However, the SV response spectrum is not explicitly provided in most of the design standards and is often approximated using the Pseudo-Spectral Velocity (PSV) spectrum. This approximation introduces substantial errors, particularly at short and long periods, necessitating more reliable PSV-to-SV conversion models. Existing conversion models largely overlook the influence of critical seismic parameters such as magnitude, source-to-site distance, site class, and other seismological parameters. This study proposes a machine-learning-based PSV to SV conversion model to integrate these parameters using a comprehensive dataset of recorded ground motions from the Engineering Strong Motion Database (ESM 2.0). The developed model is further tuned to the sparse ground motion dataset for the Indian subcontinent, leveraging a Transfer Learning framework, enabling tailored SV estimations that reflect local seismic conditions. The resulting model is observed to provide a more accurate PSV-to-SV conversion tool, optimized for the Indian seismic scenario. Further, a median and envelope intensity-based <i>SV/PSV</i> ratio spectra across seven damping ratios is proposed for inclusion in design codes. This can serve as a default reference for design applications when detailed information about the seismic scenario is unavailable.</p>