We have developed an Internet service, "PodCastle", that provides full-text searching of speech data (podcasts) on the basis of automatic speech recognition technologies. PodCastle enables a number of anonymous users to search and read podcasts, and to share the full text of speech recognition results for podcasts. However, even state-of-the-art speech recognizers cannot correctly transcribe all podcasts, because their content and recording environments vary very widely. PodCastle therefore encourages users to cooperate by correcting speech recognition errors so that podcasts can be searched more reliably. Furthermore, using the resulting corrections to train our speech recognizer, it implements a mechanism whereby the speech recognition performance is gradually improved. We started open-to-the-public experiments at http://podcastle.jp on June 12th, 2008.