In this paper, we propose a novel architecture called as the Direct-Computation-Sensing Architecture (DCSA) to directly calculates the polarization state changes caused by optical fiber vibrations with training data, offering a more accurate and responsive method than that with adaptive filter-based sensing architectures. We detected the distinct fiber vibration induced by piezoelectric ceramics in an established experimental platform, and recovered a song melody played near the optical fiber buddle from the fiber's polarization changes. We locate the source of the vibration by comparing data from both ends of a bidirectional transmission setup. Lastly, we conducted field tests under conditions involving machine-induced vibrations and natural cable movements.