EXPERIMENTS IN SPEECH DRIVEN INFORMATION RETRIEVAL FOR SPANISH LANGUAGE
César González-Ferreras, Valentín Cardeñoso-Payo · 2006
The paper reports on the evaluation of a system that al-lows users to search information using spoken queries. The front end is a large vocabulary continuous speech rec-ognizer which translates the query from speech to text and puts it through an information retrieval engine to retrieve the set of relevant documents for that query. The system is designed for Spanish language. The performance of the system was evaluated using the test suites of CLEF, which is an evaluation forum similar to TREC. 10 dif-ferent speakers were recorded reading the queries. Re-sults of different experiments are reported. Best results were obtained with a language model of 60,000 words and medium length queries: loss of precision of 23.79% (compared to using perfect transcription of the queries) with a WER of 18.4%. Overall, these results are encour-aging and provide a solid basis for the feasibility of build-ing speech driven information retrieval systems. 1.