Phonetic transcription of continuous speech databases is an important and necessary task for speech synthesis and recognition. Manual transcription requires well-trained experts and is time-consuming. So, it becomes more and more usual to implement automatic or semi-automatic transcription procedures. However, it is well known that pronunciation variations and coarticulation phenomena involve serious problems for automatic transcribers. This paper describes an approach to automatically label a continuous speech database at phonetic level using orthographic transcription (word level) and a set of pre- built acoustic models to deal with pronunciation variations. These pronunciation variations arise from coarticulation phenomena. Common intra-word and trans-word coarticulation phenomena will be described as well.