--word_timestamps = True not working

[ollisulopuisto]

Hi.

I think this might be user error (hopefully!), but when I run whisper-ctranslate2 inside Google Colab with the following command

!whisper-ctranslate2 "{full_path}" --vad_filter True --model large-v2 --verbose True --language fi --word_timestamps True --output_dir "{transcripts_dir}"

I get perfectly nice sentence-level output in JSON and every other format, but no word-level tagging. I've tried removing some of the parameters, but haven't gotten it to work yet. Removing vad_filter doesn't seem to do anything.

Is this a language-dependant thing (i.e. Finnish & large-v2) or is there something else I'm missing?

Both {full_path} and {transcripts_dir} contain meaningful values.

[jordimas]

Hello.
It's not language or model dependant. It works for me.
Some ideas:

• Make sure that you use the whisper-ctranslate2 version 0.4.1 or higher
• Look in the JSON for '"words": [' it should be there, even if is not added you should see a "words": null
• Try it outside Google Colab