We propose an approach to grapheme-to-phoneme conversion with multiple pronunciations based on a probabilistic method: Conditional Random Fields (CRF). CRF give a long term prediction and assume relaxed state independence condition compared to HMMs. Moreover, we propose an algorithm to one-to-one letter to phoneme alignment needed for CRF training. This alignment is based on discrete HMM. This paper investigated the impact of the training set size and the multiple pronunciation generation. Validated on BDLex French dictionary, our approach compares favorably with the performance of the state-of-the-art Joint-Multigram Models in term of the quality of the pronunciations and in term of recall and precision measures for multiple pronunciation variants generation.