Abstract: Seismic AVO/AVA (amplitude-versus-offset or amplitude-versus-angle) analysis, based on prestack seismic angle gathers and the Zoeppritz equation, has been widely used in seismic exploration. However, conducting the multi-parameter AVO/AVA inversion using only PP-wave angle gathers is often highly ill-posed, leading to instability and inaccuracy in the inverted elastic parameters (e.g., P- and S-wave velocities and bulk density). Seismic AVO/AVA analysis simultaneously using both PP-wave (pressure wave down, pressure wave up) and PS-wave (pressure wave down, converted shear wave up) angle gathers has proven to be an effective method for reducing reservoir interpretation ambiguity associated with using the single wave mode of PP-waves. To avoid the complex PS-wave processing, and the risks associated with PP and PS waveform alignment, we developed a method that predicts PS-wave angle gathers from PP-wave angle gathers using a deep learning algorithm—specifically, the cGAN deep learning algorithm. Our deep learning model is trained with synthetic data, demonstrating a strong fit between the predicted PS-waves and real PS-waves in a test datasets. Subsequently, the trained deep learning model is applied to actual field PP-waves, maintaining robust performance. In the field data test, the predicted PS-wave angle gather at the well location closely aligns with the synthetic PS-wave angle gather generated using reference well logs. Finally, the P- and S-wave velocities estimated from the joint PP and PS AVA inversion, based on field PP-waves and the predicted PS-waves, display a superior model fit compared to those obtained solely from the PP-wave AVA inversion using field PP-waves. Our contribution lies in firstly carrying out the joint PP and PS inversion using predicted PS waves rather than the field PS waves, which break the limit of acquiring PS-wave angle gathers.