The 1998 HTK broadcast news transcription system: development and results

Abstract

This paper presents the development of the HTK broadcast news
\ntranscription system for the November 1998 Hub4 evaluation. Relative to the previous year’s system The system a number of features
\nwere added including vocal tract length normalisation; cluster-based
\nvariance normalisation; double the quantity of acoustic training data;
\ninterpolated word level language models to combine text sources;
\nincreased broadcast news language model training data; and an extra adaptation stage using a full-variance transform. Overall these
\nchanges to the system reduced the error rate by 13% on the 1997
\nevaluation data and the final system had an overall word error rate
\nof 13.8% for the 1998 evaluation data sets.