7 Commits

Author SHA1 Message Date
Fangjun Kuang
963ac73c27 Add nn.Linear to transform the output of encoder and decoder. 2022-03-11 15:41:03 +08:00
Fangjun Kuang
ec78b7ef72 Remove extra layer norm in the conformer encoder layer. 2022-03-11 14:48:08 +08:00
Fangjun Kuang
726c92c476 Remve the last nn.Linear from the transformer model. 2022-03-11 14:21:27 +08:00
Fangjun Kuang
a5d30664bb Fix style issues. 2022-03-10 10:21:49 +08:00
Fangjun Kuang
35f5a15a54 Use giga speech dataset as extra training data. 2022-03-10 10:13:49 +08:00
Fangjun Kuang
9071b1420d Refactor decoder and joiner to remove extra nn.Linear(). 2022-03-09 23:06:48 +08:00
Fangjun Kuang
7d1b064c96 Copy files. 2022-03-09 22:36:07 +08:00