Word Level Transcripts Error

[Harinderpreet]

I have noticed that whisper x does not provide word-level transcripts for numbers. It simply mess up, it also mess up before and after words transcripts as well

[LandryBulls]

Also experiencing the same issue.

[SeeknnDestroy]

You can use supress_numerals=True in asr_options, this way model will output numbers as strings like seven and you'll be able to use world level timestamps

[Godfreyid]

@SeeknnDestroy, where can I read up on this documentation? There's a lot of commands and options to said commands, too much to try out and know what they do. How'd you figure out the 'supress_numerals=True' behavior? Do you just tinker?

[SeeknnDestroy]

@SeeknnDestroy, where can I read up on this documentation? There's a lot of commands and options to said commands, too much to try out and know what they do. How'd you figure out the 'supress_numerals=True' behavior? Do you just tinker?

I have had the same issue before, and i have searched it in the issues to learn if someone else had it too #314 (comment)

[SeeknnDestroy]

And no, i don't know if there is a documentation, just tinkering and reading the source code, searching issues etc..

[Godfreyid]

And no, i don't know if there is a documentation, just tinkering and reading the source code, searching issues etc..

Okay, thanks though.

[Godfreyid]

And no, i don't know if there is a documentation, just tinkering and reading the source code, searching issues etc..

Okay, thanks though.

I just went back and checked the anaconda terminal command: $ whisperX --help
and saw the '--suppress_numerals' option there with a brief description. So this likely is the place for your documentation.

[SeeknnDestroy]

i am using this as documentation as well if it counts https://github.com/m-bain/whisperX/blob/main/whisperx/transcribe.py#L20-L78 @GodfreyIbrahim