SpeechBot: a Speech Recognition based Audio Indexing System for the Web

We have developed an audio search engine incorporating speech recognition technology. This allows indexing of spoken documents from the World Wide Web when no transcription is available. This site indexes several talk and news radio shows covering a wide range of topics and speaking styles from a selection of public Web sites with multimedia archives. Our Web site is similar in spirit to normal Web search sites; it contains an index, not the actual multimedia content. The audio from these shows suffers in acoustic quality due to bandwidth limitations, coding, compression, and poor acoustic conditions. The shows are typically sampled at 8 kHz and transmitted, RealAudio compressed, at 6.5 kbps. Our word-error rate results using appropriately trained acoustic models show remarkable resilience to the high compression, though many factors combine to increase the average word-error rates over standard broadcast news benchmarks. We show that, even if the transcription is inaccurate, we can still achieve good retrieval performance for typical user queries (69%). Because the archive is large - over 5000 hours of content (and growing at a rate of 100 hours per week), totaling 47 million words and growing rapidly - we measure performance in terms of the precision of the top-ranked matches returned to the user.