Table 8 False alarm rates for OOV words when using Kaldi-based ASR with N-gram plus RNN re-scoring LM
From: Dynamic out-of-vocabulary word registration to language model for speech recognition
Position | Freq. | 2 mora | 3 mora | 4 mora | 5 mora | |
|---|---|---|---|---|---|---|
ONE | – | – | 0 | 0 | 0 | 0 |
ALL | – | – | .011 | .007 | .001 | 0 |
500r | Random | 500 | .010 | .011 | .001 | 0 |
1kr | Random | 1000 | .015 | .010 | .001 | .003 |
2kr | Random | 2000 | .021 | .008 | .002 | .003 |
5kr | Random | 5000 | .029 | .013 | .002 | .005 |
500c | POS | 500 | .009 | .007 | .002 | .001 |
1kc | POS | 1000 | .013 | .008 | .001 | 0 |
2kc | POS | 2000 | .019 | .010 | .002 | .001 |
5kc | POS | 5000 | .034 | .015 | .001 | .001 |