Repeat after Me: Self-Supervised Learning of Acoustic-to-Articulatory Mapping by Vocal Imitation