fix: correct segment order in batched k2 forced_align - #76
Open
lumpidu wants to merge 1 commit into
Open
Conversation
forced_align returned per-segment frame labels and scores in the wrong order for batches larger than one. encode_supervisions sorts the batch by length and returns the permutation, but forced_align never applied it, so each segment received another segment's alignment and score and the resulting durations did not match the token counts. Single-segment batches were correct because there was nothing to sort. The fix splits the scores by the sorted per-segment lengths and then restores the original batch order for both the labels and the scores. It also adds the local k2 import that forced_align was missing; without it the method raised a NameError, so the batched path had never run. The training loss is unaffected because it returns an order-independent scalar, which is why alignment training looked healthy while the exported alignments were scrambled.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
forced_alignreturned per-segment frame labels and scores in the wrong order for batches larger than one.encode_supervisionssorts the batch by length and returns the permutation, butforced_alignnever applied it, so each segment received another segment's alignment and score and the resulting durations did not match the token counts. Single-segment batches were correct because there was nothing to sort.The fix splits the scores by the sorted per-segment lengths and then restores the original batch order for both the labels and the scores.
It also adds the local k2 import that
forced_alignwas missing; without it the method raised aNameError, so the batched path had never run. The training loss is unaffected because it returns an order-independent scalar, which is why alignment training looked healthy while the exported alignments were scrambled.