Add language to word timestamps for Whisper #31572
Merged
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
_collate_word_timestamps uses the return_language flag to determine whether the language of the chunk should be added to the word's information
added missing comma
…b.com/robinderat/transformers into whisper-language-with-word-timestamps
|
Hi @robinderat, thanks for working on this! Could you please add a test in |
test that the pipeline can return both the language and timestamp
|
LGTM! thanks for iterating on this :) cc @amyeroberts for final review! |
sanchit-gandhi
approved these changes
Jul 3, 2024
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
4 tasks
|
@amyeroberts gentle reminder 😄 |
amyeroberts
reviewed
Jul 15, 2024
| @@ -1197,12 +1197,16 @@ def _find_longest_common_sequence(sequences, token_timestamp_sequences=None): | |||
| return total_sequence, [] | |||
|
|
|||
|
|
|||
| def _collate_word_timestamps(tokenizer, tokens, token_timestamps, language): | |||
| def _collate_word_timestamps(tokenizer, tokens, token_timestamps, language, return_language): | |||
There was a problem hiding this comment.
Is return_language always a bool, or could it be None too?
There was a problem hiding this comment.
It can be a bool or None. However, the behavior for return_language=None is equal to that of return_language=False
| ) | ||
| data = load_dataset("openslr/librispeech_asr", "clean", split="test", streaming=True, trust_remote_code=True) | ||
| sample = next(iter(data)) | ||
| pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language="en", task="transcribe") |
There was a problem hiding this comment.
Is this modification necessary for people to be able to use return_language with the pipeline?
There was a problem hiding this comment.
I based my test on an existing one that had this modification in it. However, it does not seem to affect the tests, so I have now removed them.
Specifying return_language=True in the pipeline is all that is required for this to work
Removed model configurations that do not influence test results
Removed model configurations that do not influence test results
amyeroberts
approved these changes
Jul 17, 2024
amyeroberts
pushed a commit
to amyeroberts/transformers
that referenced
this pull request
Jul 19, 2024
* add language to words _collate_word_timestamps uses the return_language flag to determine whether the language of the chunk should be added to the word's information * ran style checks added missing comma * add new language test test that the pipeline can return both the language and timestamp * remove model configuration in test Removed model configurations that do not influence test results * remove model configuration in test Removed model configurations that do not influence test results
MHRDYN7
pushed a commit
to MHRDYN7/transformers
that referenced
this pull request
Jul 23, 2024
* add language to words _collate_word_timestamps uses the return_language flag to determine whether the language of the chunk should be added to the word's information * ran style checks added missing comma * add new language test test that the pipeline can return both the language and timestamp * remove model configuration in test Removed model configurations that do not influence test results * remove model configuration in test Removed model configurations that do not influence test results
zucchini-nlp
pushed a commit
to zucchini-nlp/transformers
that referenced
this pull request
Jul 24, 2024
* add language to words _collate_word_timestamps uses the return_language flag to determine whether the language of the chunk should be added to the word's information * ran style checks added missing comma * add new language test test that the pipeline can return both the language and timestamp * remove model configuration in test Removed model configurations that do not influence test results * remove model configuration in test Removed model configurations that do not influence test results
itazap
pushed a commit
that referenced
this pull request
Jul 25, 2024
* add language to words _collate_word_timestamps uses the return_language flag to determine whether the language of the chunk should be added to the word's information * ran style checks added missing comma * add new language test test that the pipeline can return both the language and timestamp * remove model configuration in test Removed model configurations that do not influence test results * remove model configuration in test Removed model configurations that do not influence test results
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 language to word timestamps for Whisper
Fixes 29520
This fix enables whisper to return word-level timestamps and the predicted language at the same time. Before, enabling word-level timestamps would discard the language prediction.
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@kamilakesbi
@sanchit-gandhi