Speech enhancement guided by contextual articulatory information