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

Mistiming of lines in output #17

Open
shinchiro opened this issue Apr 4, 2015 · 5 comments
Open

Mistiming of lines in output #17

shinchiro opened this issue Apr 4, 2015 · 5 comments

Comments

@shinchiro
Copy link
Contributor

Sabagebu!-09:
http://www.solidfiles.com/d/b39659062d/Sabagebu!-09.rar
All lines in first chapter group appear late. For the next chapter groups, it give correct output

Shirobako-10:
http://www.solidfiles.com/d/eeb5b448b3/Shirobako-10.rar
I guess this is problem with search group. The typesetting lines in the end of chapter group "Part B" has moved to earlier part in chapter group "Preview" in the output

I guess this is similar problem with #16

@shinchiro shinchiro changed the title Mistiming line output Mistiming of lines in output Apr 4, 2015
tp7 added a commit that referenced this issue Apr 4, 2015
It was impossible for OpenCV to shift line starting at 00:00:00.00 back because it never changes pattern duration. With some padding it's possible to get negative shift at the beginning of the script and positive shift at its end.
Reported in #17
@tp7
Copy link
Owner

tp7 commented Apr 4, 2015

First part of the issue (sabagebu) is fixed in the commit linked above. The second part is yet another case of audio search working badly for almost silent segments and right now I don't have any real ideas on how to improve it.

By the way I noticed that you're always using --window 100 - this is not safe in general and you're getting very large mistimings exactly because of it. You should consider using the --max-window setting if you want to counter possible errors where default window is not enough, it should be more safe for general use.

@shinchiro
Copy link
Contributor Author

Thanks pointing for not using --window 100. I used Sushi.0.4.3c which doesnt yet has --max-window option. Can I get latest version with all these bugfixes?

And here is segment spectrogram of the typesetting line I reported in shirobako. Does a segment must have continous sound waves for it to work properly. Just curious :)
spectro_002

@tp7
Copy link
Owner

tp7 commented Apr 4, 2015

Get it here.

@shinchiro
Copy link
Contributor Author

thanx

@shinchiro
Copy link
Contributor Author

Dont know if this worth to mention or not, but I tried changed the small_window value to 0.5, the mistiming in the shirobako sample is completely gone. The typesetting lines appear perfectly.

Though, that small value doesnt fix completely silent segment in one of my sample

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

No branches or pull requests

2 participants