Performance Assessment of a COTS Speech Recognition System on the N4 Database
David T. Williamson, Robin A. Snyder, Jr. . · 2002
This report discusses the evaluation of a commercially available speech recognition system on the NATO Native and Non-Native (N4) database. Using the statistical language modeling techniques, trigram language models were generated for each of three countries in the database, CA, NL, and UK. Due to time constraints, the DE database was not evaluated. For each of the countries, two factors were assessed. The first was overall word accuracy and the second was callsign accuracy. For this evaluation, only standard American English acoustic models were used. Results of each country evaluation are discussed.