Vietnamese is the national and official language in Vietnam. In the Vietnamese writing system, most of vowels have diacritical signs. With this special feature, secret information can be embedded into Vietnamese documents by slightly shifting up/down and left/right these signs. The embedded documents are nearly the same as the original ones, and so the reader could not find any differences by eyes. Secret information can be extracted for many purposes such as copyright protection, tampering detection, etc. However, extracting embedded information is a really complicated task. In this paper, we propose a solution to apply Artificial Neural Network (ANN) for effectively extracting hidden information from embedded Vietnamese documents.