Post your JAV subtitle files here - JAV Subtitle Repository (JSP)★NOT A SUB REQUEST THREAD★

[ericf]

So, if I use the web version of Whisper, how can I use different models?
https://colab.research.google.com/g...in/WhisperWithVAD.ipynb#scrollTo=jh24lirNZnXT
Unless someone makes a version of Whisper as a simple installation exe with a Windows GUI, I won't be installing it. Also, on this computer I don't have a graphics card with lots of memory that I can use. It would probably take forever if what people in this thread are saying is true.

 

[r00g]

Unless someone makes a version of Whisper as a simple installation exe with a Windows GUI, I won't be installing it. Also, on this computer I don't have a graphics card with lots of memory that I can use. It would probably take forever if what people in this thread are saying is true.

Here's a Windows port: https://github.com/Const-me/Whisper

 

[ericf]

Thanks. I'll look into that. I guess I can't run this on this computer since it can't run on CPU. I'll have to try it on my W10 computer with a fairly good graphics card with GPU.

 

[ericf]

So that ffmpeg command you listed is the one they recommend:
On a bash shell I just loop through and convert all my mp4s to wavs:
for f in *.mp4; do ffmpeg -n -i "$f" -ar 16000 -ac 1 -c:a pcm_s16le "../wav/${f%.mp4}.wav"; done

I've also noticed sometimes some models work better than others, so if a movie is proving difficult I'll run a couple different models and then merge the results and clean up in aegisub.
I just pull models from here: https://huggingface.co/models?language=ja&other=whisper

Also sometimes it helps if you use an offset, so if the dialogue doesnt start for a while the transcription can be poor, so if you know the dialogue starts at 3 minutes in, then trim that off the beginning of the file, and then just offset the SRT post-processing. At least for whisper.cpp there is a parameter to do offset that makes things easier.

So how do I actually download models from HuggingFace? The links lead to pages with lots of info on the models and there's a file section but there's no .bin to download like there is here: https://huggingface.co/datasets/ggerganov/whisper.cpp/tree/main
For example: https://huggingface.co/openai/whisper-large-v2/tree/main
15 files and three of them are 6.17GB. None are bin files.
I got pointed to the WhisperDesktop program so I'm trying to find as many options for that as possible.
Thanks.

 

[counter_productive]

So how do I actually download models from HuggingFace? The links lead to pages with lots of info on the models and there's a file section but there's no .bin to download like there is here: https://huggingface.co/datasets/ggerganov/whisper.cpp/tree/main
For example: https://huggingface.co/openai/whisper-large-v2/tree/main
15 files and three of them are 6.17GB. None are bin files.
I got pointed to the WhisperDesktop program so I'm trying to find as many options for that as possible.
Thanks.

so if you don't want to use the command line and are using a colab network you can add a few lines of code to pull the model. they are very large though...

Downloading models · Hugging Face

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

huggingface.co

from huggingface_hub import hf_hub_url, cached_download
import joblib

REPO_ID = "YOUR_REPO_ID"
FILENAME = "sklearn_model.joblib"

model = joblib.load(cached_download(
hf_hub_url(REPO_ID, FILENAME)
))

 

[ericf]

Using the WhisperDesktop and the medium.bin from the link, I get a very large srt but with mostly music notes and one or two different lines of the same subtitle repeated ad nauseam. Maybe 20-30 lines of proper dialog in all.
I'm trying with the large file now. If that doesn't work, I'll just have to go on with the web version.

 

[counter_productive]

Using the WhisperDesktop and the medium.bin from the link, I get a very large srt but with mostly music notes and one or two different lines of the same subtitle repeated ad nauseam. Maybe 20-30 lines of proper dialog in all.
I'm trying with the large file now. If that doesn't work, I'll just have to go on with the web version.

When evaluating models don't worry about looking at the output from several models over a 2 hr movie. Sometimes I'll get repeating lines or music as you said, so I will have to add an offset. If using whisper.cpp you can add an offset using the '-ot' parameter. It's in milliseconds. So if dialogue doesnt start until 1min 30 sec, i'll add '-ot 90000' . Sometimes whisper also sorta drops out due to how it's looking at previous tokens, so if i notice dialogue failing later on i'll just rerun those portions but only for 20-40 minutes or whatever I need for decent coverage.

example for starting 60 minutes into a movie to grab the last half of it:
main -m models/ggml-medium-ja.bin -f "VAGU-252.wav" -ot 3600000 -d 3657000-l ja -otxt -osrt -ocsv -of VAGU-252_1hr.ja-whisper-medium

 

[ironfevers]

Using the WhisperDesktop
repeated ad nauseam. Maybe 20-30 lines of proper dialog in all.

I don't see VAD (Voice Activity Detection) in that version. It was a hassle to install https://colab.research.google.com/github/ANonEntity/WhisperWithVAD/blob/main/WhisperWithVAD.ipynb locally but it was worth it. There were so many things to install like CUDA, cudnn, (I think these two are already installed if you have used javplayer before), install https://pytorch.org, and

pip install deepl srt ffmpeg-python
pip install git+https://github.com/openai/whisper.git

I have Python 3.9.13 and no compatibility issues. I can't say if Python 3.10 and newer have issues or not with packages like numpy, numba, etc. On my local pc, I changed the code so that it can queue up multiple audio paths. I can leave my pc on when I'm not home and whisper moves on to the next file. https://drive.google.com/file/d/1z1YX-YVgTZ2LoGDxmcK3BSYnZrWpSLYg/view?usp=share_link. I will explain the changes. Global variables for all the vad settings except audio_path is a parameter. Made a function called use_whisper and indented all the previous code into it. To queue up multiple audio, just copy paste use_whisper("your_path_to_audio_file") at the end (There is commented out code as an example). Also I added more words to the garbage list. Sounds like crackling and crying, and hallucinations such as, It's in my belly, and thank you for dinner gets deleted. To run, type python whisper-vad-function.py

 

[rickson]

MIAA-750

When the subordinate knew about the Boss's adultery, she stops them from speaking by stalking them and fucking them with her big ass. "Mizuki Yayoi"​

 

[Chuckie100]

VENU-156 Miki Sato Copulation Steam Bath Incest Mother And Child​

I used Whisper to produce this subtitle file for Venu156. I love this actress and I wanted to sub this movie because it seemed to have the same actor as the mother-son couple in Venu-348. My objective was to make it a prequel to VENU-348. As always however, I still had to clean it up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese or Chinese so my re-interpretations might not be totally accurate but I try to match what is happening in the scene. Anyway, enjoy and let me know what you think of my "prequel".​

 

[Chuckie100]

I used Whisper to produce three more subtitle files for Venu-109, Venu-144, and Venu-169; which all stared Miki Satou. As always, I had to clean them up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese or Chinese so my re-interpretations might not be totally accurate but I try to match what is happening in the scene. Anyway, enjoy and let me know what you think..​

 

[Chuckie100]

FERA-118 I Started Living On My Own And My Mom Helped Me Move Into My New Apartment. After Hearing The Screams Of Sexual Pleasure Cumming From My Next Door Neighbor, She Got So Horny She Jumped On Me. Reiko Sawamura​

I used Whisper to produce this subtitle file for FERA-118. I still had to clean it up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese or Chinese so my re-interpretations might not be totally accurate but I try to match what is happening in the scene. Anyway, enjoy and let me know what you think.​

 

[kssk]

My favorite genre is incest, and I really was curious about those reality type vids where they pick up family members off the street and interview them first before leaving them alone in a room to have sex. Unfortunately those types of vids don't get translated much, probably because those have a lot of "pointless" dialogue. But I think the dialogue adds a lot to the background of the characters and the dynamics of their relationships. So I went primarily for those when playing around with Whisper and uh... the results weren't super great. Even playing around with vad and chunk thresholds still made for extremely messy translations. But at least I can get the gist of the conversations now.

I also tried some dramatic movies and Whisper works so much better for those. It still has a lot of mistakes throughout, but the first attempt is usually satisfactory for me, unlike with reality vids where I have to do multiple attempts and have to piece shit together or figure out the context myself.

Here's three completely unedited drama subtitles (DLDSS-030, DASD-862, SPRD-678) that I put through Whisper on default settings. I haven't even watched through them, but skimming through the lines they seem okay.

 

[Chuckie100]

FERA-25 Love Lust Sex Ryoko Murakami Excess Sabishinbo Mother.​

I used Whisper to produce this subtitle file for FERA-025. You gotta love Ryoko's huge boobs! I still had to clean it up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese or Chinese so my re-interpretations might not be totally accurate but I try to match what is happening in the scene. Anyway, enjoy and let me know what you think.​

 

[Taako]

VENU-156 Miki Sato Copulation Steam Bath Incest Mother And Child​

I used Whisper to produce this subtitle file for Venu156. I love this actress and I wanted to sub this movie because it seemed to have the same actor as the mother-son couple in Venu-348. My objective was to make it a prequel to VENU-348. As always however, I still had to clean it up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese or Chinese so my re-interpretations might not be totally accurate but I try to match what is happening in the scene. Anyway, enjoy and let me know what you think of my "prequel".​

You could have saved yourself some time friend, both of those been sub for years. I think 3-4 month ago I was watching them both.
Didn't care too much for Venu-348 like I last remember, the son was just too willing to get caught.
But I always appreciate your effort. Your version could be better

 

[Taako]

ASSA is the same thing as .ass btw.

Unless you want to do something more than just basic text on a video, there's really no good reason to use .ass over subrip(.srt) which is likely the most supported subtitle format.

.ass just gives you a lot of control over how the text is displayed. You can change the font of individual lines, color them, move them to a specific position, scale them bigger or smaller, add gradients to the text, fade effects, stuff like that. You can even draw shapes if you want to block some text on screen and add subtitles over that part.

Here's an example of something fancy I did back in the day, the "B&L's Style Romance!" text in that picture is 100% soft sub, you can toggle it on and off. It's 9 different subtitle lines(and possibly 2 more lines for the gradient and something else, forgot how that works 7 of which are to create the gradient effect, it only shows a slice of each line of different colors) stacked onto each other to mimic the style of the japanese title.
View attachment 3164793

Edit: One thing I forgot about that could be useful with .ass even if you only want basic text is that you can set an actor name for each line which doesn't get displayed into the subs so if you find it helpful to know who is saying what when editing, that's one thing .srt can't do.

Awesome! I have used .ass to my advantage as well

 

[Chuckie100]

You could have saved yourself some time friend, both of those been sub for years. I think 3-4 month ago I was watching them both.
Didn't care too much for Venu-348 like I last remember, the son was just too willing to get caught.
But I always appreciate your effort. Your version could be better

Yea, I had bought subtitles for VENU-348 from EroJapanese a long time ago. I especially liked VENU-348 mainly because it starred Miki Satou and Eriko Muira. However, I didn't know there were existing subtitles for VENU-156.

 

[kadal123]

when i do a whisper after 4 min processing, why a output running same until the end ??

264
00:39:14,000 --> 00:39:24,000
OK?
265
00:39:24,000 --> 00:39:32,000
OK?
266
00:39:32,000 --> 00:39:42,000
OK?

its start on 04.00 min

 

[Chuckie100]

NUKA-05 Murakami Pies Three Shots Without Disconnecting Mother And Zetsurin Son Of Frustration Ryoko​

I used Whisper to produce this subtitle file for NUKA-05, another Ryoko Murakami feature! I still had to clean it up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese or Chinese so my re-interpretations might not be totally accurate but I try to match what is happening in the scene. Anyway, enjoy and let me know what you think.​

 

[counter_productive]

when i do a whisper after 4 min processing, why a output running same until the end ??

264
00:39:14,000 --> 00:39:24,000
OK?
265
00:39:24,000 --> 00:39:32,000
OK?
266
00:39:32,000 --> 00:39:42,000
OK?

its start on 04.00 min

how are you running it? locally using python on your own GPU? With whisper.cpp? a colab instance?
So it's looking for previous tokens for context, and if there is a lot of non-normal dialogue (e.g. sex) then it can throw in a wrench in things. There are a few things to try. 1) try different models, 2) use a VAD, 3) break larger wav files into chunks and stitch them together.

Personally I've found using fine-tuned models by Japanese researchers to perform the best, and I break larger wav files into smaller ones. Some movies just perform better than others for transcription. If there are a lot of people talking, or if there isn't a lot of dialogue, or if there is a shower running or some other distracting noise I get sub-optimal results.