Add InterCTC to E-Branchformer encoder, and the ability to save InterCTC inference output to files #5084
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ate CTC predictions to file during ASR inference
for more information, see https://pre-commit.ci
|
Thanks! |
pyf98
reviewed
Apr 6, 2023
pyf98
reviewed
Apr 6, 2023
| token_list = self.beam_search.token_list | ||
|
|
||
| for layer_idx, encoder_out in intermediate_outs: | ||
| y = self.asr_model.ctc.argmax(encoder_out)[0] # batch_size = 1 |
There was a problem hiding this comment.
The intermediate CTC only supports greedy decoding? I think it's fine.
There was a problem hiding this comment.
Yes. I think greedy decoding can probably cover most use cases
|
I think it is good as long as this new feature does not affect the default behavior without interCTC. |
Codecov Report
@@ Coverage Diff @@
## master #5084 +/- ##
==========================================
- Coverage 75.88% 75.86% -0.03%
==========================================
Files 615 615
Lines 54767 54824 +57
==========================================
+ Hits 41559 41591 +32
- Misses 13208 13233 +25
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 8 files with indirect coverage changes 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
|
Thanks a lot! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add InterCTC with optional self-conditioning to E-Branchformer encoder
This was originally written by @wanchichen.
The implementation is nearly the same as InterCTC in
TransformerEncoderandConformerEncoder.Add the ability to write predictions of InterCTC layer(s) to file during inference
The output is saved at
path/to/asr_decode_dir/test_set/logdir/output.*/1best_recog/encoder_interctc_layer<layer_idx>.txt.<layer_idx>corresponds to the indices set in model config file.For example, if the config contains this line:
interctc_layer_idx: [6,12], then there will two files corresponding the output of the 6th and the 12th encoder layer:encoder_interctc_layer6.txtandencoder_interctc_layer12.txt.Misc
The code in this PR was used in many of my Aphasia speech detection experiments and produced good results. I also added more tests to make sure ASR without InterCTC still works.
@sw005320 I mentioned there was a bug I encountered in my experiments in our last meeting, but looks like it has already been patched.