Abstract

We present a system which allows the user to search for information on mobile devices using spoken natural language queries. This is the first work that we are aware of which evaluates spoken query based information retrieval on a commonly available and well researched text database, the Chinese news corpus used in National Institute of Standards and Technology (NIST)’s TREC-5 and TREC-6 conferences [TREC2001]. It is also the first work which compares the performance of spoken query information retrieval systems using 3 sets of simultaneously recorded data: headset microphone, PocketPC microphone, and cellular phone. Our result shows that while information retrieval on mobile devices with high quality microphones provides very high level of retrieval precision (e.g. mean average precision of 0.459 relative to 0.489 for searches conducted with perfect text input), the more challenging channel conditions encountered on the cellular phone require more robust and accurate acoustic models. Different types of text units were used for indexing the text database and the syllable unigram + bigram combination was found to provide best retrieval performance. 1