PodCastle: A Spoken Document Retrieval Service Improved by Anonymous User Contributions

PodCastle: A Spoken Document Retrieval Service Improved by Anonymous User Contributions Masataka Goto and Jun Ogata National Institute of Advanced Industrial Science and Technology (AIST) 1-1-1 Umezono, Tsukuba, Ibaraki 305-8568, JAPAN m.goto [at] aist.go.jp Abstract. In this invited paper, we introduce a public web service, PodCastle, that pro- vides full-text searching of speech data (Japanese podcasts) on the basis of automatic speech recognition technologies. This is an instance of our research approach, Speech Recognition Research 2.0, which is aimed at providing users with a web service based on Web 2.0 so that they can experience state-of-the-art speech recognition performance, and at promoting speech recognition technologies in cooperation with anonymous users. PodCastle enables users to find podcasts that include a search term, read full texts of their recognition results, and easily correct recognition errors by simply selecting from a list of candidates. Even if a state-of-the-art speech recognizer is used to recognize podcasts on the web, a number of errors will naturally occur. PodCastle therefore encourages users to cooperate by correcting these errors so that those podcasts can be searched more reliably. Furthermore, using the resulting corrections to train the speech recognizer, it implements a mechanism whereby the speech recognition performance is gradually improved. In our experiences from its practical use over the past 46 months (since December, 2006), we confirmed that the performance of PodCastle was improved by a number of anonymous user contributions. Keywords: information retrieval, speech recognition, error correction, wisdom of crowds, Web 2.0 1 Introduction Speech recognition researchers understand what sort of speech is easily recognized by speech rec- ognizers and realize that speech recognizers perform best when dealing with clean speech. On the other hand, most end users of speech recognizers judge the effectiveness of speech recognition from their limited experiences and do not necessarily understand how useful state-of-the-art rec- ognizers can be. Users sometimes do not adequately comprehend what sort of voices or recording conditions make recognition difficult. If they have previously had difficulty being understood by speech recognizers, they often doubt the usefulness of speech recognition and may stop using it. The first aim of this study is to address this problem by promoting the popularization and use of speech recognition by raising end user awareness of state-of-the-art speech recognition per- formance. For this purpose, we launched a podcast search web service called PodCastle (Goto et al., 2007; Ogata et al., 2007; Ogata and Goto, 2009b; Ogata and Goto, 2009a) in 2006 that allows anonymous users to search and read podcasts, and to share the full text of speech recogni- tion results for podcasts. Podcasts are audio programs distributed on the web, like radio shows or audio blogs. They are becoming increasingly popular because updated podcasts (MP3 audio files) can be easily and frequently downloaded by using RSS syndication feeds. Since various contents have already been published as podcasts, users can grasp the current state of speech recognition technology just by seeing the results of speech recognition applied to published podcasts. This is Copyright …

Content truncated to 500 words