What happens if there are multiple faces in a frame

Hi, thank you for your work! Could you please tell me what type of model you used to detect the mouth region of a person? Also, what happens if a frame contains multiple faces? Additionally, based on the mentioned images below, what type of lip frames were used to train your model? and what happens if the target speaker (with some background speakers as noise) becomes silent with lips closed in between, does the model gives silence as output during those frames?