Design of fast LVCSR systems

The paper describes the development of fast (less than 10 times real-time) large vocabulary continuous speech recognition (LVCSR) systems based on technology developed for unlimited runtime systems assembled for participation in recent DARPA/NIST LVCSR evaluations. A general system structure for 10 times real-time systems is proposed and two specific systems that have been built for broadcast news (BN) and conversational telephone speech (CTS) recognition are described. The systems were evaluated in the DARPA/NIST April 2003 rich transcription evaluation. Results are reported and contrasted with unlimited runtime systems and previous fast systems.

1. Introduction

For more than a decade a major focus in LVCSR research work have been the yearly U.S. Government sponsored evaluations conducted by NIST. While these evaluations helped the research community to accurately measure the progress in the state-of-the-art in LVCSR and led to impressive improvements in accuracy [17], they also encouraged research sites to pursue “accuracy at any price”. This lead to typical systems running in about 300 times slower than real-time (with some taking up to 2000xRT). As LVCSR technology matured there is now again an increased interest in building faster systems while retaining the gains achieved. This trend is also reflected in the recently initiated DARPA EARS programme which aims at fast transcription of both Broadcast News and Conversational Telephone Speech data.