Latency and number of model parameters of various post-processing models and the proposed CREAM framework
| Method | Model parameters | Latency | ||
|---|---|---|---|---|
| N-best reranking method | ||||
| MLM (BERT) | 102m | 0.113 | ||
| Error correction model | ||||
| BART | 140m | 0.112 | ||
| FastCorrect | 65 m | 0.014 | ||
| CREAM | ||||
| Text correction module | Text rescoring module | Text–speech matching module | ||
| BART | BERT | ASR | 272M | 1.741 |
| BART | BERT | MATE | 442M | 3.050 |
| BART | BERT | MIMICA | 344M | 1.461 |
| FastCorrect | BERT | ASR | 197M | 0.512 |
| FastCorrect | BERT | MATE | 367M | 1.797 |
| FastCorrect | BERT | MIMICA | 269M | 0.380 |
| BART | BERT | ASR+MATE+MIMICA | 574M | 3.596 |
| FastCorrect | BERT | ASR+MATE+MIMICA | 499M | 2.175 |
| Method | Model parameters | Latency | ||
|---|---|---|---|---|
| 102m | 0.113 | |||
| 140m | 0.112 | |||
| FastCorrect | 65 m | 0.014 | ||
| Text correction module | Text rescoring module | Text | ||
| 272M | 1.741 | |||
| 442M | 3.050 | |||
| 344M | 1.461 | |||
| FastCorrect | 197M | 0.512 | ||
| FastCorrect | 367M | 1.797 | ||
| FastCorrect | 269M | 0.380 | ||
| ASR+MATE+MIMICA | 574M | 3.596 | ||
| FastCorrect | ASR+MATE+MIMICA | 499M | 2.175 | |
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.