This dataset contains numerical seismo records for 1600 synthetic acoustic impedance distributions presented in [1]. Forward modeling was performed using grid-characteristic numerical method. There are nine seismo records for each impedance distribution. The records differ in excitation pulse positions. The pulses are located equidistantly on the upper boundary of the domain. Acoustic impedance distributions are in masks/ folder. For example, the file masks/1.npy contains the distribution for the sample number 1. Seismo records are in seismograms/ folder, each file name contains distribution number and pulse location. For example, the file seismograms/1_position_10.npy contains seismo record for the impedance distribution 1 obtained with the pulse center located 10 percent from the left border of the domain. The data is in NumPy native format. Total dataset size is 6.9G after extraction. Archive md5 checksum: dc9f61333f1eef17526831e0c6d9b493 [1] Fangshu Yang and Jianwei Ma (2019). Deep-learning inversion: A next-generation seismic velocity model building method. DOI: 10.1190/geo2018-0249.1