Skip to main content

Table 1 Ten hypotheses decoded from one utterance in Hub5’00-SWB set

From: Empirically combining unnormalized NNLM and back-off N-gram for fast N-best rescoring in speech recognition

 

Hypotheses

1

No no it doesn’t work on every issue

2

No no it doesn’t work in every issue

3

No no it doesn’t work and every issue

4

No no it doesn’t work out every issue

5

Oh no no it doesn’t work on every issue

6

No no it doesn’t work then every issue

7

No no he doesn’t work on every issue

8

Oh no no it doesn’t work in every issue

9

On no no it doesn’t work on every issue

10

No no he doesn’t work in every issue

  1. Lots of contexts are similar and differences usually exist in local.