I prefer to comment on the video as a whole, which I regard as a stunning achievement. Farrar's speaking voice was well-recorded in the 1930s, and her "cloned voice" conveys her well-documented assertiveness. She also looks as she did during the years in which she was doing intermission commentaries during Metropolitan Opera broadcasts. When the image changes to one taken and AI-animated from a photograph from the period in which she recorded "Serenata," we hear AND see her singing. The fact that the software is still under refinement and momentarily exaggerates mouth movements is for me a minor issue in the context of the entire video.
The program is labeled "Beta" meaning "don't expect perfection." When I realize the technology that made even this level possible I find it remarkable.