The statistical pattern recognition is a promising framework for text-to-text translation. However, a natural extension to speech-input translation is not straightforward. In this paper, we present a method to deal with the speech input statistical translation problem that could be considered as a step towards a fully integrated recognition-translation procedure. In this version, a word graph was used in the input as a representation of the acoustic of a given utterance. As a case study, experimental results with the so-called "Traveler task" are presented by using a text-input statistical translator.