We have proposed a language model adaptation method for captioning disaster broadcasts. In this paper, we develop a prototype system to collect training texts for the adaptation. Through experiments on speech recognition, the effectiveness of our system was evaluated.
