A novel two-level architecture plus confidence measures for a keyword spotting system

Autor: Tejedor Noguerales, Javier, King, Simon, Frankel, Joe, Wang, Dong, Colás, José, Garrido, Javier
Přispěvatelé: UAM. Departamento de Ingeniería Informática, Laboratorio de Tecnología Hombre-Computador (ING EPS-010)
Jazyk: angličtina
Rok vydání: 2009
Předmět:
Zdroj: Tejedor, J, King, S, Frankel, J, Wang, D, Colas, J & Garrido, J 2009, A novel two-level architecture plus confidence measures for a keyword spotting system. in Proceedings of the 5th Biennial Workshop on Speech Technology . pp. 247-250 .
Biblos-e Archivo. Repositorio Institucional de la UAM
instname
Popis: This is an electronic version of the paper presented at the V Jornadas en Tecnología del Habla, held in 2008 on Bilbao
In this work, we present a novel two-level architecture for a keyword spotting system. The first level is composed of an HMM-based keyword spotting process. The second level uses isolated word recognition. Two confidence measures in the decision stage, based on the posteriors and the keywords hypothesised by this second level, are presented and compared within the keyword spotting system. Both confidence measures outperform the performance of the first level in isolation.
This work was partly funded by the Spanish Ministry of Science and Education (TIN 2005- 06885).
Databáze: OpenAIRE