Inference: pad very short signals before embedding them #14055
Merged
Conversation
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
The speaker embedding model crashes on very short signals. So we zero-pad the end of the signal if it's less than 0.5 seconds long before running it through the speaker embedding model. Signed-off-by: Fejgin, Roy <[email protected]>
subhankar-ghosh
approved these changes
Jul 1, 2025
| def extract_embedding(model, extractor, audio_path, device, sv_model_type): | ||
| speech_array, sampling_rate = librosa.load(audio_path, sr=16000) | ||
|
|
||
| # pad to 0.5 seconds as the extractor may not be able to handle very short signals | ||
| speech_array = pad_audio_to_min_length(speech_array, int(sampling_rate), min_seconds=0.5) |
There was a problem hiding this comment.
Has this been tested, does this affect the final evaluation metrics in any way?
There was a problem hiding this comment.
I tested that the padding works correctly. Did not collect pre/post evaluation stats. This should only kick in very rarely, when the generated speech is shorter than 0.5 sec.
There was a problem hiding this comment.
Following up on this: I ran libri_unseen_test with and without the padding fix and found no statistically significant differences in WER and SSIM.
Thanks @subhankar-ghosh. It had been set to auto-merge so already got merged as soon as you approved, but I'll still look at your comments. |
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Outdated suggestions cannot be applied.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.
The speaker embedding model crashes on very short signals. So we zero-pad the end of the signal if it's less than 0.5 seconds long before running it through the speaker embedding model.