In this project, we aim to investigate how foreign accents, ethnicity, gender, and speech-accompanying gestures influence investors' decisions in the entrepreneurs' investment pitch. To experimentally manipulate ethnicity while retaining the paralinguistic information from speech and gestures, we will employ deep learning face swapping algorithm in creating the stimuli. Specifically, we will swap the face of the stimuli created by an ethnic majority actor with the face of an ethnic minority actor. This is possible by utilizing the head model in the DeepFaceLab. The official examples are as below:

We also generated the demo with limited computing resources (2080TI for 48 hours). This demo utilized a whole face model but not a head model, so the forehead is not included. Future work will employ the head model and manually adjust the forehead's color tone by programs such as PhotoShop and AfterEffect.

To manipulate gender while retaining the paralinguistic information from speech and gestures. Besides DeepFaceLab, we will employ the voice-conversion algorithm, which enables us to convert the speech of a male to a female and vice versa. Below are the demos: