New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add E-Branchformer configs and models in ASR recipes #4837
Conversation
Codecov Report
@@ Coverage Diff @@
## master #4837 +/- ##
=======================================
Coverage 79.18% 79.18%
=======================================
Files 557 557
Lines 49279 49279
=======================================
Hits 39020 39020
Misses 10259 10259
Flags with carried forward coverage won't be shown. Click here to find out more. 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
Hi @sw005320 Can we merge this PR now (before SLT) so that some people can start to use E-Branchformer? I will continue to add more in following PRs. |
OK, I'll start to review. |
Thanks @sw005320 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM.
I have minor comments.
Thanks a lot! |
Following #4833, this PR continues to add E-Branchformer results in ESPnet2 ASR recipes. The updated recipes and results (including those in the previous PR) are summarized below. Joint CTC/attention training and decoding are performed unless otherwise specified. In ESPnet, the most widely used Conformer config is:
12 layers, 2048 linear units (8 times expansion)
.aidatatang_200zh: CER, with Transformer LM
aishell: CER, without LM
chime4: WER on beamformit_5mics, with Transformer LM
gigaspeech: WER, without LM
#4882
librispeech_100: WER, without LM
librispeech_100: WER, CTC only, beam size 1, without LM
swbd: WER, without LM
tedlium2: WER, without LM
tedlium2: WER, CTC only, beam size 1, without LM
voxforge: CER, without LM
wsj: WER, with Transformer LM