Increasing machining efficiency using intelligent methods is a crucial objective in Industry 4.0. Using Deep Learning methods for detection and avoidance of machining chatter has been becoming popular recently. One of the drawbacks of deep learning methods is their requirement for large amounts of data for training to perform with sufficient accuracy. Moreover, the data should be balanced within the classes. The collection of chatter data for training classifiers is an expensive and laborious task. This study investigates the use of Variational Auto Encoders (VAE) for synthetic data generation to improve the performance of Convolutional Neural Networks (CNN). A limited amount of data is collected from an experimental setup and transformed into Hilbert-Huang transform (HHT) images. Afterward, the number of images is increased significantly, eliminating the imbalance between stable and chatter classes. The performance of several CNN classifiers using synthetic data is evaluated, and future research directions are suggested.