Table 7 Perplexities of language models trained from mixed-style and spoken-style utterances evaluated on VT-TST and SOC sets
From: Classification-based spoken text selection for LVCSR language modeling
LM | PPL | ||||
|---|---|---|---|---|---|
Base | 146.71 | ||||
ALL | 144.74 | ||||
LM (174K) | PPL | LM (151K) | PPL | LM (98K) | PPL |
Random.174K | 148.36 | Random.151K | 148.45 | Random.98K | 148.47 |
PPL.174K | 144.91 | PPL.151K | 144.38 | PPL.98K | 139.56 |
CRF | 144.08 | LSTM | 140.45 | SVM | 139.18 |