History

Hui Zhang e0a87a5ab1 batch average ctc loss (#567 ) * when loss div batchsize, change lr, more epoch, loss can reduce more and cer lower than before * since loss reduce more when loss div batchsize, less lm alpha can be better. * less lm alpha, more cer reduce * alpha 2.2, cer 0.077478 * alpha 1.9, cer 0.077249 * large librispeech lr for batch_average ctc loss * since loss reduce and model more confidence, then less lm alpha		3 years ago
..
conf	batch average ctc loss (#567 )	3 years ago
local	batch average ctc loss (#567 )	3 years ago
.gitignore	Support paddle 2.x (#538 )	3 years ago
README.md	batch average ctc loss (#567 )	3 years ago
path.sh	Support paddle 2.x (#538 )	3 years ago
run.sh	Refactor CTC module, add embedding and fix log (#549 )	3 years ago

LibriSpeech

CTC

Model	Config	Test Set	WER	Valid Loss
DeepSpeech2	conf/deepspeech2.yaml	test-clean	0.069357	15.078561
DeepSpeech2	release 1.8.5	test-clean	0.074939	15.351633