One-shot Talking Face Generation from Single-speaker Audio-Visual Correlation Learning