I did a test and got 1800 lines, it is possible that your source file is corrupted, to generate this test use the supjav version.
Post your JAV subtitle files here - JAV Subtitle Repository (JSP)★NOT A SUB REQUEST THREAD★
- Thread starter Eastboyza
- Start date
-
Throughout the month of April 2024, participate in the FileJoker Thread Contest OPEN TO EVERYONE!
From 1st to 30th of April 2024, members can earn cash rewards by posting Filejoker-Exclusive threads in the Direct-Downloads subforums.
There are $1000 in prizes, and the top prize is $450!
For the full rules and how to enter, check out the thread -
Akiba-Online is sponsored by FileJoker.
FileJoker is a required filehost for all new posts and content replies in the Direct Downloads subforums.
Failure to include FileJoker links for Direct Download posts will result in deletion of your posts or worse.
For more information see this thread.
Thanks for the tip! I used supjav, converted to a sound file via VLC, and ran it again. Made it to 1:07:00 this time - 1000 lines. Its strange because the converted file still has clear sound but Whisper/VAD process did nothing with it. I'll mess around with it again and let you know. I don't want to clutter up the jav subtitle posting area with my tech problems too much.
If your internet service is not very good or at least 50 MB, I do not recommend viewing or downloading files from the internet when using Whisper.Thanks for the tip! I used supjav, converted to a sound file via VLC, and ran it again. Made it to 1:07:00 this time - 1000 lines. Its strange because the converted file still has clear sound but Whisper/VAD process did nothing with it. I'll mess around with it again and let you know. I don't want to clutter up the jav subtitle posting area with my tech problems too much.
a situation that I have detected, is the use of my computer's ram, using between 10 and 11 MB, this happens when I am using Whisper in Chrome (Mention that I have up to 8 pages open and downloading AV videos)
I second @soloporhoy666 note, it probably is caused because of a audio decoding error. What does ffprobe say? If you share your audio file I can try to run a test on it.Thanks for the tip! I used supjav, converted to a sound file via VLC, and ran it again. Made it to 1:07:00 this time - 1000 lines. Its strange because the converted file still has clear sound but Whisper/VAD process did nothing with it. I'll mess around with it again and let you know. I don't want to clutter up the jav subtitle posting area with my tech problems too much.
I have gigabit speed, and I don't download anything concurrently that may interrupt it. Its just this one video (for me obviously) from multiple sources that is the weird part! Running ffprobe appears beyond my expertise... The zipped file is 150 mb which appears to large for the server. I'll just break it up into chunks for Whisper and merge when done.., but thanks for the offer. Maybe it's a VLC thing? Hmmmm.
Well, if you want it, later he published it here, edit it and upload it again, it looks good, just because he loved Kira, that athletic and well marked body drives me crazy.
A Boxing Day present for Aoi Tsukasa fans: here is sub for SSIS-581.
Aside to @Taako, can I ask for a favor? Would you spot check this transcription for the level of accuracy/errors? I enjoyed reading your earlier posts and your review of an earlier sub. I digged into Whisper a bit more and tweked the settings to find closest match for the time accuracy and dialogue detection. Would be helpful if someone that knows the language could check the final sub. Caveat: for the first scene, I have manually addedd the translation for the written Questions, and correction of he/she. The rest is untouched raw from the machine.
Just another note: the sub is based on 30fps version of the movie [720p]. The 1080p 29.75 fps 1080p version of the movie has audio decoder errors in couple of places. You may need to adjust the fps if needed.
Aside to @Taako, can I ask for a favor? Would you spot check this transcription for the level of accuracy/errors? I enjoyed reading your earlier posts and your review of an earlier sub. I digged into Whisper a bit more and tweked the settings to find closest match for the time accuracy and dialogue detection. Would be helpful if someone that knows the language could check the final sub. Caveat: for the first scene, I have manually addedd the translation for the written Questions, and correction of he/she. The rest is untouched raw from the machine.
Just another note: the sub is based on 30fps version of the movie [720p]. The 1080p 29.75 fps 1080p version of the movie has audio decoder errors in couple of places. You may need to adjust the fps if needed.
Attachments
Last edited:
I know a little of the language and will do my best. I haven't sub anything of Aoi, so I can't compare anything I might have laying around. But I'm sure you did great!
From what Solo said, the more people who uses Whisper, the better it becomes.
Please make sure the mp3 and mp4 files have the same time to limit errors when using Whisper.
I don't think it is the size, I have uploaded files that exceed 200 mb, they also exceed 2 hours, so the error must be something else.
In case someone is interested in editing it, even though it turned out quite well for me, some words but in general quite decent, but if someone has knowledge and time, maybe they can give us an improved version of these subtitles.
codeED-020 AIKA
code
ED-020 AIKAHey fellow subbers,
Been following the Whisper+VAD stuff and I gotta say this stuff is 100% better then any process for subbing that I've used in the past. My question today is are there any tips for improving time using the google collab version? The subbing takes around 25-45 minutes per video and I was wondering if this is the standard or is there something I can do to improve that?
Also does anyone know if I audio boost the MP3 files before going in, would that help with audio detection? I have noticed some dialogue that is spoken quietly is not picked up.
PS. Ill be dropping some subs Ive done to give back to the community, keep an eye out for that.
Been following the Whisper+VAD stuff and I gotta say this stuff is 100% better then any process for subbing that I've used in the past. My question today is are there any tips for improving time using the google collab version? The subbing takes around 25-45 minutes per video and I was wondering if this is the standard or is there something I can do to improve that?
Also does anyone know if I audio boost the MP3 files before going in, would that help with audio detection? I have noticed some dialogue that is spoken quietly is not picked up.
PS. Ill be dropping some subs Ive done to give back to the community, keep an eye out for that.
I envy those of you who can use Whisper to make subtitles, seems complicated to install but it also seems much faster than my method. My method is very archaic, I'm learning Japanese (I'm already at an advanced stage of my learning process) and I do everything by hand just listening to what they're saying and I only use machine translation when there's really loud music or when the actresses are practically whispering.
That's why is kinda slow, but very accurate. I'm maybe 35% done with MIAE-208 which is a really, really dialogue heavy long movie.
That's why is kinda slow, but very accurate. I'm maybe 35% done with MIAE-208 which is a really, really dialogue heavy long movie.
From what I have learned so far:
- Translate is faster than transcribe task -- that is done by design by the OpenAI developers;
- The free version of colab has no guarantee on time --shared resources situation; 25-45min is a good range;
- On a pro colab T4 continues to be a cost optimal GPU;
- On a pro colab, with an A100 GPU one can get 2x speed --half time of T4, but the cost is 4x;
- I suaully avoid volume boost for the entire movie -- it distorts other parts;
- For scenes that have quient whipsering sounds I cut them, boost volume and then feed them separately;
Good luck and happy subbing
Have you tried the collab version? It's real easy to use, nothing you have to do other then download the movie and click buttons there is zero coding and you dont need to manually install anything. A user posted an easy to follow guide on this thread and I followed it first time with ease.
In some of the features that I mentioned in another post, I added that whisper could help with the timing of a movie, by generating more lines of dialogue than the other systems, you could use this to your advantage and you don't need to install anything with the virtual version ( ok, yes you have to install the repository that the creator of the virtual collab provides to be able to use whisper)
Google Colaboratory
colab.research.google.com
Google Colaboratory
colab.research.google.com
Use the virtual version, easy, just follow the instructions, this collab is easy, drag your mp3, wait for it to load, copy the address of the uploaded file, install the two options that the collab asks for each time you use virtual whisper and ready you have created your subtitle.