Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Detect common transcript errors #7

Open
waldoj opened this issue Mar 12, 2023 · 3 comments
Open

Detect common transcript errors #7

waldoj opened this issue Mar 12, 2023 · 3 comments

Comments

@waldoj
Copy link
Member

waldoj commented Mar 12, 2023

I'm seeing Whisper do goofy, incorrect stuff sometimes.

One is the insertion of earlier phrases later on in transcripts, sprinkled through the transcript.

Another is repeating a phrase dozens of times:

[00:04:05.000 --> 00:04:08.000]   >> A question whether that is necessary for an objection
[00:04:08.000 --> 00:04:11.000]   to the question.
[00:04:11.000 --> 00:04:14.000]   >> I think that's a good question. I think that's a good
[00:04:14.000 --> 00:04:17.000]   question. I think that's a good question.
[00:04:17.000 --> 00:04:20.000]   >> I think that's a good question. I think that's a good
[00:04:20.000 --> 00:04:23.000]   question. I think that's a good question. I think that's a
[00:04:23.000 --> 00:04:26.000]   good question. I think that's a good question. I think that's
[00:04:26.000 --> 00:04:29.000]   a good question. I think that's a good question. I think that's
[00:04:29.000 --> 00:04:32.000]   a good question.
[00:04:32.000 --> 00:04:35.000]   >> I think that's a good question. I think that's a good
[00:04:35.000 --> 00:04:38.000]   question. I think that's a good question.

Identify a lightweight way to identify these artifacts in transcripts and flag them for review or re-processing.

@waldoj
Copy link
Member Author

waldoj commented Mar 12, 2023

Well, it ain't fancy, but this would identify the presence of often-repeating lines:

grep -E ".{31,}" transcript.srt |sort |uniq -c |sort -nr |head -1 |xargs |cut -d " " -f 1

waldoj added a commit that referenced this issue Mar 15, 2023
Identifies duplicate lines. Toward #7.
@waldoj
Copy link
Member Author

waldoj commented Mar 15, 2023

Now figure out what to do with these, once detected. My best guess is this:

  • Store the list in a file
  • Use the file as a blacklist for future transcribing
  • Remove references to the transcripts from the JSON, if present
  • Surface this list via some mechanism, so that transcripts aren't just silently, permanently missing

@waldoj
Copy link
Member Author

waldoj commented Mar 15, 2023

OK, they're now stored in a file, and used to prevent adding new references to the JSON. I think I should just manually prune existing JSON references.

It won't be possible to properly close this issue until I understand the source of this problem (why is Whisper doing this?) and how to fix it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant