Spoken Document Retrieval in a Highly Inflectional Language
Inger Ekman, Kalervo Järvelin · DSpace repository (University of Tartu) · 2007
Being able to search for relevant information within a collection of documents is vital for the effective use of any kind of storage media.The main body of Spoken Document Retrieval research has concentrated on the English language, leaving other languages -and information produced in them -without the benefit of existing SDR systems.Especially the performance of vocabulary based speech recognition suffers from inflection.This also affects the usability of retrieval systems based on vocabulary-based recognition techniques.We discuss a method for rapid phoneme filtering to facilitate fast searches for words unknown by large vocabulary speech recognizers.The method is evaluated against a speech database in Finnish, which is a highly inflected language.