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

[Chuckie100]

VENU-617 Relatives [silence] Gonna Have A Dad In Incest Next To ... Ayumi Shinoda​

I recently downloaded what I thought was a reduced mosaic version of VENU-613 but what I got was a reduced mosaic version of VENU-617 starring Ayumi Shinoda my GOAT of JAV actresses! I used WhisperJAV0.7 to update a previous Sub that I had posted and I also attempted to clean it up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese 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.

The movie had a lot of action but little dialog; however, it did have a lot of Ayumi!.

 

[ironfevers]

It's been a long time since I posted. I recently discovered how good Deepseek is, and I got excited. I spent ~30 hours on this Python Script to translate Chinese SRT files to English using Deepseek v3. Why use it? My Python script translates in batches, so it has the context of previous lines. Create an account on fireworks ai. Click on your profile picture and get your API KEY. It comes with $1 free credit. Check usage in billing. $1 can translate at least 100 subtitles. I wouldn't top up as it's not the cheapest. I plan to convert the Python script to use the official Deepseek API which is cheaper but I'll wait for my ChatGPT o3 quota to refresh before coding again. I couldn't make more subs as I spent all my credit testing the same subs over and over again for any differences when I change the system prompt and temperature values.

Set your API KEY in powershell and restart the terminal. I put some instructions at the top of the script but if you get stuck just ask AI. The script is ready to use! I put all the best default settings. I find temperature 0.9 to be the upper limit for explicit dirty language before responses are likely to ignore the rules and format. If you find translations mismatching with the timecode, then lower temperature by 0.1 each run. Top_p 0.95 is good. SYS_MSG is already good but if you want to change the erotic instructions then that's ok. BATCH_SIZE_DEFAULT 500 is a good upper limit. Too high and it will exceed the max tokens, too low and it will make more API calls, increasing costs. If you need to change source language to Japanese, then rewrite the SYS_MSG and rewrite the example.

I'll talk about the challenges and how the code works in case anyone wants to play with the code. Fixing bugs of problematic responses and testing took majority of the time. "SEQ_NUMBER\nORIGINAL_TEXT <eol>\n" is how I send it to get translated. I omit the timecode to save tokens. Before this I tried a <space> in between the number and text but it wasn't a robust structure and I encountered line merges all the time, so do not do this. The <eol> is a safety net that reinforces the structure. When temperature is too high, the responses are less likely to follow the rules. Problematic responses given back can contain ">" after the number, the incorrect number, the English translation will merge two Chinese lines and translate as one (this is the biggest problem which causes timecode mismatch when combining), and other quirky oddities. I accounted for nearly all this in the code. Map_translated function parses the response and cleans it up. Suspected short and missing lines get sent to the prompt again for re-translation. If you need to fix bugs, the log files are useful for checking responses. Things I wanted to test but got lazy: testing different values of top_p between 0.96 to 1 with different values of temperature, and testing the code without inserting <eol>. If we truly don't need <eol>, then we can save a few tokens each line, but I'm not 100% sure if it's useful in preventing line merges or not. Anyway, you're free to do whatever you want with the code.

Subs of Mori Hinako (favorite actress), Akai Miki, ROE-168 (mother-son), MIAA-750 (female slutty boss, amazing loud plopping cowgirl), YUJ-031 (high energy, enthusiastic girl that kisses a lot) and more!

 

[ironfevers]

Just learned about this. Use the new model for improved translations. Change from MODEL = "accounts/fireworks/models/deepseek-v3" to MODEL = "accounts/fireworks/models/deepseek-v3-0324". My first run caused line merges at TEMPERATURE = 0.9. My new recommended default is TEMPERATURE = 0.8

Nao Jinguji, she is getting thick. Beautiful jiggly ass!

 

[DScott]

Stay tuned for revised info

 

[DScott]

It's been a long time since I posted. I recently discovered how good Deepseek is, and I got excited. I spent ~30 hours on this Python Script to translate Chinese SRT files to English using Deepseek v3. Why use it? My Python script translates in batches, so it has the context of previous lines. Create an account on fireworks ai. Click on your profile picture and get your API KEY. It comes with $1 free credit. Check usage in billing. $1 can translate at least 100 subtitles. I wouldn't top up as it's not the cheapest. I plan to convert the Python script to use the official Deepseek API which is cheaper but I'll wait for my ChatGPT o3 quota to refresh before coding again. I couldn't make more subs as I spent all my credit testing the same subs over and over again for any differences when I change the system prompt and temperature values.

Set your API KEY in powershell and restart the terminal. I put some instructions at the top of the script but if you get stuck just ask AI. The script is ready to use! I put all the best default settings. I find temperature 0.9 to be the upper limit for explicit dirty language before responses are likely to ignore the rules and format. If you find translations mismatching with the timecode, then lower temperature by 0.1 each run. Top_p 0.95 is good. SYS_MSG is already good but if you want to change the erotic instructions then that's ok. BATCH_SIZE_DEFAULT 500 is a good upper limit. Too high and it will exceed the max tokens, too low and it will make more API calls, increasing costs. If you need to change source language to Japanese, then rewrite the SYS_MSG and rewrite the example.

I'll talk about the challenges and how the code works in case anyone wants to play with the code. Fixing bugs of problematic responses and testing took majority of the time. "SEQ_NUMBER\nORIGINAL_TEXT <eol>\n" is how I send it to get translated. I omit the timecode to save tokens. Before this I tried a <space> in between the number and text but it wasn't a robust structure and I encountered line merges all the time, so do not do this. The <eol> is a safety net that reinforces the structure. When temperature is too high, the responses are less likely to follow the rules. Problematic responses given back can contain ">" after the number, the incorrect number, the English translation will merge two Chinese lines and translate as one (this is the biggest problem which causes timecode mismatch when combining), and other quirky oddities. I accounted for nearly all this in the code. Map_translated function parses the response and cleans it up. Suspected short and missing lines get sent to the prompt again for re-translation. If you need to fix bugs, the log files are useful for checking responses. Things I wanted to test but got lazy: testing different values of top_p between 0.96 to 1 with different values of temperature, and testing the code without inserting <eol>. If we truly don't need <eol>, then we can save a few tokens each line, but I'm not 100% sure if it's useful in preventing line merges or not. Anyway, you're free to do whatever you want with the code.

Subs of Mori Hinako (favorite actress), Akai Miki, ROE-168 (mother-son), MIAA-750 (female slutty boss, amazing loud plopping cowgirl), YUJ-031 (high energy, enthusiastic girl that kisses a lot) and more!
View attachment 3670033View attachment 3670034View attachment 3670035View attachment 3670036View attachment 3670037

IStay tuned for revised info

 

[mei2]

1500 .srt files

.... I welcome any tweaking suggestions.

Thanks for the collection. I plan to check them out during the weekend.
Meanwhile I noticed you're not using --word_timestamps. That should give you more accurate timing.
Also, you're using --task translate with model large-v3. That model was not trained for translation task over large-v2. It was only additionally trained for --task transcribe.

 

[DScott]

Thanks for the collection. I plan to check them out during the weekend.
Meanwhile I noticed you're not using --word_timestamps. That should give you more accurate timing.
Also, you're using --task translate with model large-v3. That model was not trained for translation task over large-v2. It was only additionally trained for --task transcribe.

Mei, thank you for this input. I have been wrestling back and forth with V2 and V3. I had originally be advised by a reliable source that V2 was actually superior to V3 with respect to Japanese-English translation. Later the same source said , V3 is superior. My intent with these subs is to upload them for Akiba members and consequently I want to have the most usable/accurate translations. I installed the faster-whisper model but I have never been able to get it to work. I honestly would prefer to use V2 because it is quicker but that said my own 'research' seems to indicate V3 is better. Just an example, in one file that i tested the interviewer in the beginning was referring to Gokkun. IN V2 it was translated as Kokkun while V3 translated it as Gokkun. Now I'm sure not many people wouldn't have had a problem making the connection with V2's Kokkun but again I am after the most perfect translation that i can find. I'll add the Word timestamps arg to my batch but could you please clarify for me about V2 vs. V3 because if I can get away with V2 and speed up the process that would be super-keano. Thanks.

 

[aronora]

VENU-617 Relatives [silence] Gonna Have A Dad In Incest Next To ... Ayumi Shinoda​

View attachment 3669937
I recently downloaded what I thought was a reduced mosaic version of VENU-613 but what I got was a reduced mosaic version of VENU-617 starring Ayumi Shinoda my GOAT of JAV actresses! I used WhisperJAV0.7 to update a previous Sub that I had posted and I also attempted to clean it up a bit and re-interpreted some of the meaningless/ "lewd-less" dialog. Again, I don't understand Japanese 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.

The movie had a lot of action but little dialog; however, it did have a lot of Ayumi!.

About contents of that stepmother/stepson secret sex - next to son's father (and remarried mature woman/stepmother's husband).
Actress - Ayumi Shinoda - who was active at Japanese porn industry on 2010, and after first retirement (during that was made a significant boobs upgrade) returned in 2014 and retired again on 2016. Currently 44 years old, but not performing any more already 9 years. In particular movie - VENU-617 she is in real life 36 years old.

Stepson - actor Oyoyo Nakano , started his career in Japanese porn industry on 2006 when 20 years old. In VENU-617 he is in real life 30 years old.

Son's father and Ayumi's remarried husband - actor Naoki Otsuka - close to 50 years old.


Introductory text for VENU-617 from FANZA page -

"My new remarried husband has a son from a previous marriage, called Tohru.
He seems to be at a difficult age, and even now, after six months our marriage,
Tohru still won't talk to me properly. I didn't think it would be that easy to become
a parent and child, but I wanted us to at least have a fun chat...
Then, one day, after getting out of the bath, I encountered Tohru looking through
my underwear in the dressing room.
Surprised, I confronted Tohru, but he said something I never expected..."

Just to explain about real contents of this movie - if you don't understand Japanese...
This is movie about stepmother/stepson intercourse, no blood connection.

Of course, one could imagine/dream about real incest, watching whatever exciting movie.

 

[t221152]

Mei, thank you for this input. I have been wrestling back and forth with V2 and V3. I had originally be advised by a reliable source that V2 was actually superior to V3 with respect to Japanese-English translation. Later the same source said , V3 is superior. My intent with these subs is to upload them for Akiba members and consequently I want to have the most usable/accurate translations. I installed the faster-whisper model but I have never been able to get it to work. I honestly would prefer to use V2 because it is quicker but that said my own 'research' seems to indicate V3 is better. Just an example, in one file that i tested the interviewer in the beginning was referring to Gokkun. IN V2 it was translated as Kokkun while V3 translated it as Gokkun. Now I'm sure not many people wouldn't have had a problem making the connection with V2's Kokkun but again I am after the most perfect translation that i can find. I'll add the Word timestamps arg to my batch but could you please clarify for me about V2 vs. V3 because if I can get away with V2 and speed up the process that would be super-keano. Thanks.

I've done some tests as well and V2 is just better from the tests I've done. While there were a few tests where V3 was better, in general V2 was the clear winner for me. From what I've read is that V3 is only better if the audio is super clear, like a podcast. In JAV the audio level quality can vary a lot.

I also have better results with silero_v4_fw vs pyannote_v3 (VAD method). mdx_kim2 (voice extraction method) also doesn't work super great for me, but can help in splitting different speakers I've noticed. But mdx is not in faster_whisper I think ?

There's 1 specific JAV that is really bad for me, and I've tried so many different things on it but I can barely get anything out of it. It's WAAA-501. The audio is super clear as well. There's 1 subtitle on subtitlecat which looks like it has been done with whisperjav and it's having the exact same problem.

 

[DScott]

I've done some tests as well and V2 is just better from the tests I've done. While there were a few tests where V3 was better, in general V2 was the clear winner for me.

From what I've read is that V3 is only better if the audio is super clear, like a podcast. In JAV the audio level quality can vary a lot.

I'm in the early stages of converting my Second round. I think that I'll re-set and run it with V2 for about a Week and see what I come up with. Thanks for the input T22.

 

[Chuckie100]

Of course, one could imagine/dream about real incest, watching whatever exciting movie.

I guess that is one advantage of not knowing the Japanese lanuage!

 

[DScott]

I've done some tests as well and V2 is just better from the tests I've done. While there were a few tests where V3 was better, in general V2 was the clear winner for me. From what I've read is that V3 is only better if the audio is super clear, like a podcast. In JAV the audio level quality can vary a lot.

I also have better results with silero_v4_fw vs pyannote_v3 (VAD method). mdx_kim2 (voice extraction method) also doesn't work super great for me, but can help in splitting different speakers I've noticed. But mdx is not in faster_whisper I think ?

There's 1 specific JAV that is really bad for me, and I've tried so many different things on it but I can barely get anything out of it. It's WAAA-501. The audio is super clear as well. There's 1 subtitle on subtitlecat which looks like it has been done with whisperjav and it's having the exact same problem.

Well I just spent the last couple of Hours goofing around with WAAA-501. IT is very very weird. It is almost like there's a second audio track that his coded into the file and Whisper is decoding that. I tried V2 and V3, I tried upscaling the video to see if the re-encoding may resolve the issue. I'm flummoxed! The only thing I can suggest is have someone here manually create a subtitle file. There was a cat doing it a few Months ago but I cannot remember his name. I wish that i could have found the solution but no dice here.

 

[t221152]

Yep, I've tried to re-encode it as well, to no avail. It's very weird

 

[Chuckie100]

Well I just spent the last couple of Hours goofing around with WAAA-501. IT is very very weird. It is almost like there's a second audio track that his coded into the file and Whisper is decoding that. I tried V2 and V3, I tried upscaling the video to see if the re-encoding may resolve the issue. I'm flummoxed! The only thing I can suggest is have someone here manually create a subtitle file. There was a cat doing it a few Months ago but I cannot remember his name. I wish that i could have found the solution but no dice here.

I stripped the audio into a mp3 file as normal from a reduced mosaic version of WAAA-501 and created this Sub with WhisperJAV0.7. Although some missing dialog but looks pretty normal Whisper Sub to me. Attached is the "raw" Sub with no editing.

PS: After watching the video a bit, I noticed some of the text has a very short duration time and is not seen in real time while watching the movie with the Sub. Suggest increasing the min duration limit to at least 2 seconds.

 

[t221152]

I stripped the audio into a mp3 file as normal from a reduced mosaic version of WAAA-501 and created this Sub with WhisperJAV0.7. Although some missing dialog but looks pretty normal Whisper Sub to me. Attached is the "raw" Sub with no editing.

PS: After watching the video a bit, I noticed some of the text has a very short duration time and is not seen in real time while watching the movie with the Sub. Suggest increasing the min duration limit to at least 2 seconds.

There's a lot of dialogue in the first half hour, you have 3 lines in the first half hour.

The file is only 10kb as well, I got 18kb worth of text out of it. It's how I immediately noticed something was off as well. I've downloaded the 4k version just now, see if anything changes but I doubt it, it's in the queue, will take a bit.
I've tried different VAD's on it (with super low VAD sensitivity) and also a different voice extraction method, it didnt change much. Will try it again on the the 4k version

 

[mei2]

I am after the most perfect translation that i can find. I'll add the Word timestamps arg to my batch but could you please clarify for me about V2 vs. V3 because if I can get away with V2 and speed up the process that would be super-keano. Thanks.

We peobbaly need to take this conversation to another thread so we don't hi-jack the main theme. The work that you're planning to do (and doing) is quite valuable and deserves more attention. I will spend more time on it during the weekend, but I thought of doing a quick exploration into your first set of the subs by running some stats on them. Here are some interesting stats:


(1) Hallucination: Your de-halluicnation seems to have worked quite well for known phrases. There doesn't seem to be too many of those occuring in the majority of the files:








(2) Quality: Looking at the characters per seconds (CPS) metric can give an indication of the quality of the timing, or the sub. That can be one way of measuring the quality of a sub. I'd say anything below 10 and higher than 20 might be not optimal.







(3) Quality: Another metric I looked at was for repetition. One metric can be Type-Token Ratio (TTR). This is a bit difficult metric for JAV (or any porn movie), as the vocabulary is not that vast However, one doesn't want the TTR to be too close to 0. That can indicate problems in Whisper output like repetition loop.





These are just dumb stats with no knowledge of the genre, or type of movies. So it must be taken with grain of salt. It would be interesting to run the stats for each separate genre or series to see what are the characteristics.

 

[Chuckie100]

There's a lot of dialogue in the first half hour, you have 3 lines in the first half hour.

The file is only 10kb as well, I got 18kb worth of text out of it. It's how I immediately noticed something was off as well. I've downloaded the 4k version just now, see if anything changes but I doubt it, it's in the queue, will take a bit.
I've tried different VAD's on it (with super low VAD sensitivity) and also a different voice extraction method, it didnt change much. Will try it again on the the 4k version

You are right...there is definitely something squirrely about WAAA-501! Think I have wasted enough time on it too! lol

 

[Electromog]

Not sure what's going on with WAAA-501, but I tried both the decensored version and a version with hardcoded Chinese subs and they had the same issue. 3 lines of text and then nothing more for a very long time.

So either there's something weird going on with the original soud, or the various different versions out there are all from the same corrupted source.

 

[mei2]

...there is definitely something squirrely about WAAA-501

You're right. It seems that there is something very wiered about its audio coding. When the audio is converted to mono 16KHz audio, it looses almost all its waveform (see below). That is the cause of the problem with subtitling --Whipser uses mono 16KHz wav audio.


Original waveform:



After mono resampling:




I am now trying to find a workaround. Will update later.

 

[Chuckie100]

You're right. It seems that there is something very wiered about its audio coding. When the audio is converted to mono 16KHz audio, it looses almost all its waveform (see below). That is the cause of the problem with subtitling --Whipser uses mono 16KHz wav audio.


Original waveform:
View attachment 3671857


After mono resampling:

View attachment 3671856


I am now trying to find a workaround. Will update later.

Maybe the JAV studios have found a way to screw Whisper Subs!?