Speech Transcription and Spoken Document Retrieval in Finnish

This paper presents a baseline spoken document retrieval system in Finnish that is based on unlimited vocabulary continuous speech recognition. Due to its agglutinative structure, Finnish speech can not be adequately transcribed using the standard large vocabulary continuous speech recognition approaches. The definition of a sufficient lexicon and the training of the statistical language models are difficult, because the words appear transformed by many inflections and compounds. In this work we apply the recently developed language model that enables n-gram models of morpheme-like subword units discovered in an unsupervised manner. In addition to word-based indexing, we also propose an indexing based on the subword units provided directly by our speech recognizer, and a combination of the both. In an initial evaluation of newsreading in Finnish, we obtained a fairly low recognition error rate and average document retrieval precisions close to what can be obtained from human reference transcripts.