Comparison of GANs for Speech to Face Generation Using a Custom Indian Face Dataset

An efficient face-generation technique that finds importance in forensic applications is proposed in this paper. The technique mainly aims at sketching out suspects with high accuracy making the work of police personnel simpler. Here, our focus is to present a user-friendly technique by leveraging the significant features of Generative Adversarial Networks (GANs). Audio description provided by the witness is fed as input to GANs to generate accurate and real face images. The main idea behind our work is to streamline the process of facial sketching to identify the crime suspect during investigation. Facial images from various angles are recorded and used during the face-generation process, which increases the efficacy of criminal identification process. This application will also assist in enhancing public safety and law enforcement.