Spoken query processing (SQP) is the task of fulfilling an information need, inferred from a spoken query, by listing a set of ranked relevant documents. The two main sources of uncertainty in SQP lay on the realization of the speech waveform and on the realization of the observed document. The proposed integration models these uncertainties under a single probabilistic framework. A case study on movie title retrieval by voice is presented to illustrate the proposed methodology. By allowing an ontology inlet, a 14% relative gain in the model convergence was achieved. An improved mean reciprocal rank and mean inclusion rate of the retrieval outcome was obtained using the proposed framework.
Bibliographic reference. Moreno-Daniel, A. / Wilpon, J. / Juang, B.-H. / Parthasarathy, S. (2008): "Towards the integration of automatic speech recognition and information retrieval for spoken query processing", In INTERSPEECH-2008, 2154-2157.