@shinmera matched filter approach based on fragments of the person you expect to hear talking talking?