ICASSP
2010
IEEE
13 years 11 months ago
2010
IEEE
This paper addresses the problem of discriminative training of language models that does not require any transcribed acoustic data. We propose to minimize the conditional entropy ...