Table 8.

Performance comparison between baseline models and CREAM variants under the low-error condition. The text correction and text–speech matching modules can be implemented with different settings, while the text rescoring module is implemented using BERT

MethodLow-error scenario
AISHELL-1TEDLIUM-2Libri-otherLibri-clean
Baselines
Top-17.229.727.212.58
Oracle4.146.284.451.34
BART7.5310.667.423.05
FastCorrect6.81
BERT5.558.446.862.73
CREAM
Text correction modulTextspeech matching module
BARTASR5.308.437.212.89
BARTMATE5.709.637.393.48
BARTMIMICA5.799.807.282.95
FastCorrectASR5.09
FastCorrectMATE5.21
FastCorrectMIMICA5.29
BARTASR+MATE+MIMICA4.988.136.722.70
FastCorrectASR+MATE+MIMICA4.73

or Create an Account

Close subscription notice
Close access options