Disable strict hotword matching mode for offline transducer #1837
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
…roduces new variable, so that later this mode can be switched on in the runtime.
|
@pkufool Please have a review. |
|
@vsd-vector I would suggest to use non_strict for hotwords. So I think make strict mode false by default is enough, don't necessary to make it configurable. |
|
@pkufool I removed the variable and replaced it with "false" constant value. |
# for free
to join this conversation on GitHub.
Already have an account?
# 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Outdated suggestions cannot be applied.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.
I noticed that for offline transducer strict_mode is True when traversing ContextGraph, but for online non-strict mode is used (see #638 ).
What is the rationale behind this?
In my testing, when using strict_mode the decoder sometimes "gets stuck" in the final state of the hotword graph, penalizing all tokens after the hotword. In other words, if we have sentence "he is my friend" and hotwords list contains "he", then sometimes decoder output is just "he" and remainder of the sentence is truncated. This is typically happens if weight is too large, but I would still consider such behaviour undesirable for ASR (not keyword spotter).
This PR proposes to disable strict hotword matching mode for offline transducer.
Also it introduces new variable, so that later this mode can be made configurable.