Vocabulary-independent word confidence measure using subword features

• DOI: 10.21437/icslp.1998-817
• Publication Date: 2022-08-01T16:26:52Z
• Authors: Li Jiang; Xuedong Huang
• Keywords: Discriminative model; Word error rate

This paper discusses how to compute word-level confidence measures based on sub-word features for large-vocabulary speaker-independent speech recognition. The performance of measure using at word, phone and senone level is experimentally studied. A framework transformation function system proposed high estimation. In this system, discriminative training used optimize the parameters function. comparison baseline, experiments show that reduces equal error rate by 15%, with up 40% false acceptance reduction various fixed rejection rate. combination multiple under also discussed.