Neural Rescoring#1986
Closed
Adel-Moumen wants to merge 35 commits into
Closed
Conversation
13 tasks
Merged
Adel-Moumen
changed the base branch from
ctc-prefix-beamsearch
to
unstable-v0.6
…echbrain into neural-rescoring
Collaborator
Author
|
Closing this PR in favour of #2187 |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR implements neural rescoring which can ranks the top-K candidates produced by the beam search decoding.
The interface lets you use any LMs that you wants. For instance, in general when someone wants to train an ASR system he has to make sure that the tokens used by the system are the same than the pretained LM. However, in this PR, you can take a BPE LM and use it with a CHAR system. No need to match the same granularity!
Todo:
20 beam size, alpha = 0.6 and beta = 0.25
7.30% test clean with TransformerLM from our sb hub
8.41% test clean with GPT2-medium
9.38% test clean with no lm
Note 1: this PR do not leverage #1596 as it is unnecessary
Note 2: this PR should be merged after the PR on pyctcdecode / CUDACTC / framectc so that I can integrate the n-best rescoring for them.