An active speaker detection model to detect which people are speaking in a video.

video

the start time of the video to process (in seconds)

start_time

the end time of the video to process (in seconds). If -1, process until the end of the video.

end_time

whether to return the visualization of the video.

return_visualization

a string of face boxes in the format "frame_number,x1,y1,x2,y2,x1,y1,x2,y2,..." separated by new lines per frame. If not provided, the model will detect the faces in the video itself to then detect the active speaker.

face_boxes

the maximum number of frames to load in memory at once. can speed up processing. if 0, this feature is disabled.

in_memory_threshold

if return_visualization is True, the first element of the tuple is the output of the model, and the second element is the visualization of the video. Otherwise, the first element is the output of the model.

TalkNet-ASD