The Akiba-online English Sub Project★NOT A SUB REQUEST THREAD★

[maload]

Is there anyone who can do it for a fee? I really need one episode from this collection

there is " translators " here ... his login

https://www.akiba-online.com/members/darksider59.976341/

infact if you look at the front page of the
https://www.akiba-online.com/thread...sub-request-thread.12173/page-57#post-4533693

there is " resident japanese " , or people who have skill in japanese in the first or in the begining of the topic .
i am here when i meet them,,, but i dont know that they are here now...

i only see darksider around here sometime ...

https://www.akiba-online.com/members/darksider59.976341/

 

[Taako]

Is there anyone who can do it for a fee? I really need one episode from this collection

Doubtful, you will have your request done for free. Subbing is hard work, even a 30 min scene can have lots of dialogue. It's not easy.

You should try to commission it(pay for it) or learn to do it yourself, or wait for a Chinese sub. Good luck

 

[Mabok]

Ladies (There maybe one or two out there, you never know) and gentlemen:

In the world of JAV there has always been...



View attachment 2374425 View attachment 2374426 View attachment 2374427
Big,..................................BIGGER..........................BIGGGEST (So far....)



No, not BOOBIES! Subtitle collections! A few years ago the collection of 5,500 Chinese subtitle appeared, then a few months ago @
Arny Jacksonposted 11,500+ subtitle pack. Today I post the..

28,000+ Subtitle pack!​

Yes, that's right , more that 28,000 Chinese subtitles of JAV covering the past 2 Decades (2001-2021)
These titles are ready for translation with you favorite Machine Language translation scheme (I like using Subtitle Edit with google translate) But Please, Please, take some time afterwards to proof read and make he language more nature (it really takes you out of the moment when the lovely lady comments on how big the guys "Meat Stick" is. )
After you have cleaned up the subtitles, please post them for the rest of us to enjoy, thanks.

Now 28,000 subtitles is a lot to dig though so to make life easier I did two things:

1. I translated all of the directories and sub directories from Chinese into English

2. I scraped the list of files against the movie database at JAV Library, It matched almost 16,000 Films (most of th unmatched 12,000 films are from smaller studios of website based production, This sadly includes uncensored products by 1Pondo and Caribbean), The 16,000 found films were placed into an Excel spread sheet containing English translations of the film title, Actress names, Genres, studio, director and A COVER PICTURE! This excel file can be filter and sorted to find the movies you want.
Like Ayumi Shinoda (who doesn't?)? Type in her name, Boom! 140 films
View attachment 2374430


- titles movie code and cover photo. Only want Ayumi Shinoda's cosplay films? and that and Boom - 2 listings (wish there were more...).

View attachment 2374431

Someone else was looking for Hypnosis films, found 93 of those!

View attachment 2374432

You get the idea, It's a big file but I strongly advise you to check it out and let me know what you think!

I want to thank @ironfevers for helping me to retrieve this sub pack and @theydonotwantto for creating the software I used to create the searchable, Filterable Excel file!

The subpack is here : (330MB compressed, over 1 GB uncompressed:
https://mega.nz/file/3IMBDYxT#3v_pFm1zqypEVRUtbGHYVudUMK3CaUkvSUZdTWtKCKo


The Excel files are here,
NEW VERSIONS!:
I'm releasing 2 new versions of the excel file, both are Unprotected (no password) which will make it easier to do searches, just be careful not to save the file when you exits excel (it will ask if you want to save) or yo risk screwing up the file.

The first version is the LITE version, it's doesn't contain the cover images so it's a very small file, you can do the same searched but will have to click on individual links to see each cover picture.:

LITE (3.63 MB):
https://mega.nz/file/OUkTgIZK#TjgLcGX45jNvnXYN-MH5zVO-TMa1_N6MfeD_e0_8Oyo

The second version is, is the UNLIMITED Version, it's the same big excel file containing all of the cover pics so you can see them alll at once after a search, just unlocked to make searches work easier!

UNLIMITED (530MB):
https://mega.nz/file/SMtTCQYR#FRl-qIj08S3bvesA33-xY_TUavUaiVZfSQwEsxMGRS0


PLEASE READ FOR IMPORTANT INSTRUCTIONS:
The number of images was causing problems when searches, sorting and filtering was going on, I added the ablity to hide the picture column (the document now loads with that column hidden, to un-hide it click on the ' + ' symbol above the ACTRESS column, once it's un-hidden click on the ' - ' symbol above the ACTRESS column, to re-hide it. If you are simply going to browse and scroll through the list, you can do that with the pictures un-hidden.

If you are doing searches or filtering do this:

1. Hide the pictures
2. Do your search, filter or sort
3. Once the results are up, un-hide the pictures
4. Before you clear the search, start a new search, or add more criteria to the current search, hide the picture column again.
5. Go to step 2.



Please let me know what you think!

Sorry bro, but I can't open this file

 

[SamKook]

Sorry bro, but I can't open this file

You realize there's more than 1 file in that post and there's nothing anyone can do to help you out unless you give us more details on what the issue you're having is and what you tried/ are trying to achieve.

The subpack and the unlimited version of the excel datasheet works fine for me and I'm sure many other people.

 

[Pendekar2020]

Sorry bro, but I can't open this file

Or download this. It includes the 28,000+ subtitles and more

File folder on MEGA

mega.nz

Note: it's not mine, I got it from ScanLover forum.

 

[lnstar]

Or download this. It includes the 28,000+ subtitles and more

File folder on MEGA

mega.nz

Note: it's not mine, I got it from ScanLover forum.

The file has been removed.

 

[IdeNali]

There is a new library from OpenAI "whisper" that can be found here to translate and transcribe: https://github.com/openai/whisper

You need to be a little bit tech savvy to use it but the results are really promising and runs locally on your PC without the need of any online services. I'm currently playing around with it and as far as it seems to me the results are better than with any other autosub tool that can be found online. I think one of the current limiting factors is the bad audio quality of some JAVs.

I think with some manual tune ups after the transcription/translation the results could be very good. I still need to try out the large language model that should produce the bests results but I probably don't have enough VRAM for that.

 

[SamKook]

That whisper software seems pretty good at first glance, first time I see a result that looks acceptable to me. It does seem to repeat a line when it can't understand what is being said though or it's a bug. Haven't had much luck with the medium model though(after 45 min, it just repeat the same line for the rest of the 3 hour video), had to use the large one to get something good, but even using large it'll fail on content it succeeded on before so who knows.

It's also very easy to use, or I'm just too tech savvy to see the difficulty in it, lol.


To install it, you need to have python(3.10 is what I used) and pip installed first.
If you're on windows, you need to do

pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113/

first if you have an nvidia gpu so it can use cuda which is going to be tons faster than only using the cpu.

If you installed it already and didn't do the above, you'll need to uninstall and reinstall those 3 modules to get the cuda version so this to uninstall and the line above to reinstall:

pip3 uninstall torch torchvision torchaudio

Then you do what they actually say you need to do to install it which is

pip install git+https://github.com/openai/whisper.git

To use it, you demux the audio track(seems to work if you feed it the video directly), open a windows command line and do something like this:

whisper SSIS-381.aac --model large --language ja --task translate

and it will create a vtt subtitle file once it's done, simple as that.

If you don't have enough VRAM for the model you wanna use(it should give some kind of out of memory error if you don't), you can also use the cpu to do it by adding "--device cpu" in there and if you have enough normal RAM free, it'll be able to get things done, but very slowly.

 

[maload]

If you don't have enough VRAM for the model you wanna use(it should give some kind of out of memory error if you don't), you can also use the cpu to do it by adding "--device cpu" in there and if you have enough normal RAM free, it'll be able to get things done, but very slowly.

i have 2 gig ram but its very old graphic card (gt 730 )

the program load the model and such ( 100 %) but in the end my pc is freeze
can you guide how some good enough graphic card for this ?
if i can buy it so i can upgrade my pc

 

[SamKook]

i have 2 gig ram but its very old graphic card (gt 730 )

the program load the model and such ( 100 %) but in the end my pc is freeze
can you guide how some good enough graphic card for this ?
if i can buy it so i can upgrade my pc

You should be ok using up to the small model with only 2GB, but that's probably not ideal for the result. Although the 730 is quite old and I don't think it supports recent versions of cuda so that's likely why your pc is freezing.
A bit of research tells me the latest drivers that has working cuda for it is 388.71 so you'd need to downgrade to that and use an older version of cuda for pytorch, maybe something like "https://download.pytorch.org/whl/cu80"(instead of the cu113 one in the pip install command) but I'm just guessing here.

For buying, the requirement for something like this would be an nvidia gpu so you can use cuda(the dev did say other things might work, but they only tested cuda and cpu, not opencl or stuff like that which would work on AMD cards) and a gpu with at least 10GB of VRAM if you want to use the best model. The more cuda cores the faster it'll be. I have an RTX 3080 10GB but it's far from cheap, even though price have stabilized. A 3060(not the ti one since it only has 8GB) with 12GB would be much cheaper and work fine too, but it has less than half the cuda cores so it's more than half as slow, but it's also half the price.
It would probably be a good idea to wait till mid-october when the 4000 series card get released, price might go down or something better might come up.

With that said, it doesn't seem very reliable, every time I run it on the same file I get a wildly different result, even if I use the exact same settings. The parts that work are fairly decent though from what I watched.

 

[maload]

You should be ok using up to the small model with only 2GB, but that's probably not ideal for the result. Although the 730 is quite old and I don't think it supports recent versions of cuda so that's likely why your pc is freezing.
A bit of research tells me the latest drivers that has working cuda for it is 388.71 so you'd need to downgrade to that and use an older version of cuda for pytorch, maybe something like "https://download.pytorch.org/whl/cu80"(instead of the cu113 one in the pip install command) but I'm just guessing here.

For buying, the requirement for something like this would be an nvidia gpu so you can use cuda(the dev did say other things might work, but they only tested cuda and cpu, not opencl or stuff like that which would work on AMD cards) and a gpu with at least 10GB of VRAM if you want to use the best model. The more cuda cores the faster it'll be. I have an RTX 3080 10GB but it's far from cheap, even though price have stabilized. A 3060(not the ti one since it only has 8GB) with 12GB would be much cheaper and work fine too, but it has less than half the cuda cores so it's more than half as slow, but it's also half the price.
It would probably be a good idea to wait till mid-october when the 4000 series card get released, price might go down or something better might come up.

With that said, it doesn't seem very reliable, every time I run it on the same file I get a wildly different result, even if I use the exact same settings. The parts that work are fairly decent though from what I watched.

thank you for the guiding.

 

[Taako]

That whisper software seems pretty good at first glance, first time I see a result that looks acceptable to me. It does seem to repeat a line when it can't understand what is being said though or it's a bug. Haven't had much luck with the medium model though(after 45 min, it just repeat the same line for the rest of the 3 hour video), had to use the large one to get something good, but even using large it'll fail on content it succeeded on before so who knows.

It's also very easy to use, or I'm just too tech savvy to see the difficulty in it, lol.


To install it, you need to have python(3.10 is what I used) and pip installed first.
If you're on windows, you need to do

pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113/

first if you have an nvidia gpu so it can use cuda which is going to be tons faster than only using the cpu.

If you installed it already and didn't do the above, you'll need to uninstall and reinstall those 3 modules to get the cuda version so this to uninstall and the line above to reinstall:

pip3 uninstall torch torchvision torchaudio

The you do what they actually say you need to do to install it which is

pip install git+https://github.com/openai/whisper.git

To use it, you demux the audio track(seems to work if you feed it the video directly), open a windows command line and do something like this:

whisper SSIS-381.aac --model large --language ja --task translate

and it will create a vtt subtitle file once it's done, simple as that.

If you don't have enough VRAM for the model you wanna use(it should give some kind of out of memory error if you don't), you can also use the cpu to do it by adding "--device cpu" in there and if you have enough normal RAM free, it'll be able to get things done, but very slowly.

Thank you for running a "test" and giving us the result.
It doesn't sound like it worth having right now... did you try a movie under 3 hours?
I usually subs movies under 2 hours. The older JAVs are usually under 3 hours. The new ones are usually 2 and more hours.
Maybe the WhisperAI do better with smaller movies? Anyway thank you taking the time to experiment with it.

I'll stick with Audacity for audio tweaks, pytranscriber for my timing codes and some translations, and my friends on the net

 

[SamKook]

Thank you for running a "test" and giving us the result.
It doesn't sound like it worth having right now... did you try a movie under 3 hours?
I usually subs movies under 2 hours. The older JAVs are usually under 3 hours. The new ones are usually 2 and more hours.
Maybe the WhisperAI do better with smaller movies? Anyway thank you taking the time to experiment with it.

I'll stick with Audacity for audio tweaks, pytranscriber for my timing codes and some translations, and my friends on the net

I tried chopping the movie in 45 mins segments and it didn't change anything, was just as random.

Trying it with the cpu atm to see what the result is compared to the full 4 cuda runs I did, but that's gonna take 2-3 days to complete vs 2-3 hours with the GPU, never actually properly timed it.

Attached is an example of the results you get and how it varies.

 

[Taako]

I tried chopping the movie in 45 mins segments and it didn't change anything, was just as random.

Trying it with the cpu atm to see what the result is compared to the full 4 cuda runs I did, but that's gonna take 2-3 days to complete vs 2-3 hours with the GPU, never actually properly timed it.

Attached is an example of the results you get and how it varies.

Yeah I see the results. Doesn't seem worth it right now.
I'll stick with my old way
Thank you.

 

[quay2]

https://mega.nz/file/iIkklDLY#3Lhz9IE41BEjwSLBfnGHjZVtkUvaifKpKOHyn2hN-z8 (V8) is down, btw.

 

[maload]

i cant believe i still dont know how to delete my own comment 00

 

[SamKook]

Don't think you can with this new forum.

 

[maload]

Don't think you can with this new forum.

ok, i got it

 

[panop857]

Whisper seems really impressive. The Large model is a lot to ask and very few people will have a PC capable of it, but the Small model is pretty reasonable and gets pretty good results that can be edited. Medium better still, and there's a few adapted models that claim to have better Japanese performance, like this one:

vumichien/whisper-medium-jp · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

I'm new and haven't figured out how to run it yet.

Whisper is the future of JAV subs. The floodgates are opening. do people have general suggestions? no_speech_ threshhold = 0.3 and logprob_threshhold=0.1 seem to be a sweet spot for me. What other settings do people suggest for JAV?

 

[SamKook]

It is impressive but also very random with the results so it's more of a good starting point than a full solution in my opinion, but the result is good enough for many.

There's a few collab options that allow people without a top class GPU to use the bigger models like the one mentioned in the other subtitle thread on the forum, here's the last mention of it: https://www.akiba-online.com/thread...not-a-sub-request-thread.1466451/post-4632709
That thread has been talking about whisper stuff for the past few months so lots to read on that over there.

I have no idea how to run those custom models, I've only been doing tests to help others since my initial batch of testing to see how well it worked so you just taught me about them.

Finding the best settings is hard since the randomness of the result mean you never know if the settings gave you a good/bad result or it's just random you got a good/bad one.