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

[porgate55555]

Here is a machine generated sub for Yua Mikami SSIS-477. I'm using a new Japanese app --it looks promising so far. Better than Autosub and PyTranscribe, but still not perfect.

View attachment 3005563

Looks pretty good. Do you mind sharing the name of the app you are using?

 

[kadal123]

Here is a machine generated sub for Yua Mikami SSIS-477. I'm using a new Japanese app --it looks promising so far. Better than Autosub and PyTranscribe, but still not perfect.

View attachment 3005563

what apps's name ? tell me please

 

[mei2]

Looks pretty good. Do you mind sharing the name of the app you are using?

Of course, here is the app: https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-japanese
It is rather technical / complicated at this point.

 

[mei2]

what apps's name ? tell me please

Here it is (funny I just responded to above text before seeing this one): https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-japanese

 

[maload]

Here is a machine generated sub for Yua Mikami SSIS-477. I'm using a new Japanese app --it looks promising so far. Better than Autosub and PyTranscribe, but still not perfect.

View attachment 3005563

can you give me japanese text version of this .srt please
i want to read it

 

[lnstar]

thank you.

I'm also a huge AIKA fan.

 

[ericf]

Thanks for posting that links. I don't use that kind of script language a lot. Do I just copy the section up top that begins with:
"from huggingsound import SpeechRecognitionModel"
and input the full input/output paths to the audio in the command prompt and it will do its work?
I don't see any info on having to download any files.

Also, it says it needs input of 16 kHz sample rate. So you would need to convert every audio file in Audacity to 16 kHz.
Is there a runtime limit to the audio? Vrew has a 2 hr limit and I know pyTranscriber doesn't like long files either.

Once again, thank you for letting us know there are more options out there.
I do remember someone else mentioning a 'new' method of extracting text from audio in THIS thread. Never found a more detailed explanation.

 

[mei2]

can you give me japanese text version of this .srt please
i want to read it

Here you go, here is the Japanese text.
PS. the timing is off --there was a error in split and merge recalculation.

 

[mei2]

Thanks for posting that links. I don't use that kind of script language a lot. Do I just copy the section up top that begins with:
"from huggingsound import SpeechRecognitionModel"
and input the full input/output paths to the audio in the command prompt and it will do its work?
I don't see any info on having to download any files.

Also, it says it needs input of 16 kHz sample rate. So you would need to convert every audio file in Audacity to 16 kHz.
Is there a runtime limit to the audio? Vrew has a 2 hr limit and I know pyTranscriber doesn't like long files either.

Once again, thank you for letting us know there are more options out there.
I do remember someone else mentioning a 'new' method of extracting text from audio in THIS thread. Never found a more detailed explanation.

I presume you aready have your python environemnt set up, then you'd only need to install and import the modules.

Both the input and output need pre and post work. And you're right, using Audacity for 16khz would do the job. I use ffmpeg for the same. I cut the input to 30min chunks. I'm still looking for a pre-trained (Japanese) tokenizer for preprocessing.

For pyTranscriber I usually split the input to 5min chunks before feeding it to pyTransciber.

PS. I also noticed that the results are better when audio is converted from streo to mono.
PPS. Somewhere I read that Vrew gives better results if the audio is converted to female voice, and the volume is increased 5x. I haven't tried it though.

 

[ericf]

I have used Vrew a little and it finds more dialog than pyTranscriber (autosub) but it's not always more accurate. I thought it was more accurate with Korean since it uses a dedicated Korean api but I guess Google is superior in this case.
How would one convert audio to female voice? Higher pitch? I know that amplifying works only to a certain degree with Google API/Autosub and then it drops a lot of the dialog. It just doesn't recognize it. I've always wondered why that didn't work so well. I can't do any adjustments in the API, so doing adjustments to the audio should help. I've tried compression as well. I tend to do at least 3 versions of the audio and then I compare and choose the best versions and copy missing lines between scripts.
With mono tracks, are you saying that works better with pyTranscrier? I've tried a few times but have not had any difference between stereo and mono audio.

 

[mei2]

I have used Vrew a little and it finds more dialog than pyTranscriber (autosub) but it's not always more accurate. I thought it was more accurate with Korean since it uses a dedicated Korean api but I guess Google is superior in this case.
How would one convert audio to female voice? Higher pitch? I know that amplifying works only to a certain degree with Google API/Autosub and then it drops a lot of the dialog. It just doesn't recognize it. I've always wondered why that didn't work so well. I can't do any adjustments in the API, so doing adjustments to the audio should help. I've tried compression as well. I tend to do at least 3 versions of the audio and then I compare and choose the best versions and copy missing lines between scripts.
With mono tracks, are you saying that works better with pyTranscrier? I've tried a few times but have not had any difference between stereo and mono audio.

I haven't tried the voice gender conversion --it is indeed to shift the pitch, I understand. Here is a discussion that seems to point on Audacity plugin:

https://www.reddit.com/r/kdenlive/comments/jnt4vf
I'm impressed with you method to create 3 versions -- well done!

I will give Vrew a try. I wonder if it has a capability to learn/improve. I read that Vrew records the users activities. I wonder if that is for the purpose of improving the algorithm.

 

[maload]

Here you go, here is the Japanese text.
PS. the timing is off --there was a error in split and merge recalculation.

thank you

 

[ericf]

Just a quick update: Don't use downsampled Mono with Vrew. I've only checked the Korean language extractor, but if you make it mono and 16Hz you'll get roughly a third less text in the file. I tried a mono file at 48Hz and it worked almost as well as an unadjusted stereo track. Amplify it in parts and you might catch some more dialog but the program tends to interpret amped content differently. Also, I tried a compressed file (Threshold -16db, Noise Floor: -35 db) and didn't see any real improvement.
All attempts at removing all sounds but the dialog have resulted in sound with a changed character. Either muffled or 'glitchy' and clipped. There is background noise some of the time but the most difficult parts are when there's music that's as loud as the dialog.

 

[Taako]

Looks like someone is taking subs here and posting to their site or what ever. And they do not give credit to the person who created the sub(s) lol

My recent sub of GG-016 was found here, https://javluna1.com/jav-english-su...-during-a-visit-to-my-parents-momoka-nishina/

And if the person had took the sub had read the summary, they would know...
I post only the short version here https://www.akiba-online.com/thread...not-a-sub-request-thread.1466451/post-4552511.

NOT the full sub. Which I decided not too lol

And the admin there say he created it hahaha

 

[maload]

Looks like someone is taking subs here and posting to their site or what ever. And they do not give credit to the person who created the sub(s) lol

My recent sub of GG-016 was found here, https://javluna1.com/jav-english-su...-during-a-visit-to-my-parents-momoka-nishina/

And if the person had took the sub had read the summary, they would know...
I post only the short version here https://www.akiba-online.com/thread...not-a-sub-request-thread.1466451/post-4552511.

View attachment 3007847View attachment 3007848View attachment 3007849NOT the full sub. Which I decided not too lol

And the admin there say he created it hahaha

i do hardsub and after i share it on torrent community for 1 day
all my hardsub have been found on 2 websites
funny as hell, they make money from my hard work.
there is seriouse talking about this in my little forum.
somebody said when i share it on the net so then its free for eveybody.
5 years ago , i make subtitle from english subtitle ,
its not free , we hired people who know japanese for making subtitles.
i do it for my friends in that forum.
how come they said its free for everybody.
nowsday...
its worst.
somebody just uoload thing on the net because they can make money from it
4 posts for 1 dollar and such .
i dont think this people are websites owner.
i guess they dont even watch what they upload on the net because they dont have time for that .
thing as,,, they are sitting here and that forum .... looking at that topic every 3 hour and got some subtitle ....

 

[Taako]

i do hardsub and after i share it on torrent community for 1 day
all my hardsub have been found on 2 websites
funny as hell, they make money from my hard work.
there is seriouse talking about this in my little forum.
somebody said when i share it on the net so then its free for eveybody.
5 years ago , i make subtitle from english subtitle ,
its not free , we hired people who know japanese for making subtitles.
i do it for my friends in that forum.
how come they said its free for everybody.
nowsday...
its worst.
somebody just uoload thing on the net because they can make money from it
4 posts for 1 dollar and such .
i dont think this people are websites owner.
i guess they dont even watch what they upload on the net because they dont have time for that .
thing as,,, they are sitting here and that forum .... looking at that topic every 3 hour and got some subtitle ....

Yes. So somebody here is taking subs and trying to make money here. I don't care but I won't help them.

That's why I don't share full subs anymore.
I only share subs that I clean up, find, and I ALWAYS give credit when I can.

My personal subs only go to people I know and trust

 

[Imscully]

Looks like someone is taking subs here and posting to their site or what ever. And they do not give credit to the person who created the sub(s) lol

My recent sub of GG-016 was found here, https://javluna1.com/jav-english-su...-during-a-visit-to-my-parents-momoka-nishina/

And if the person had took the sub had read the summary, they would know...
I post only the short version here https://www.akiba-online.com/thread...not-a-sub-request-thread.1466451/post-4552511.

View attachment 3007847View attachment 3007848View attachment 3007849NOT the full sub. Which I decided not too lol

And the admin there say he created it hahaha

I have become increasingly disturbed by sites that steal others work without even giving credit or reference to the original author.

 

[Chuckie100]

I found machine subs for VENU-790 starring Chisato Shouda. The sub was pretty good for a machine translation. I cleaned it up a bit and tried to interrupt what was being said on some lines, but otherwise did not try to twist the story-line. Anyway enjoy and let me know what you think.

 

[Taako]

I have become increasingly disturbed by sites that steal others work without even giving credit or reference to the original author.

Yes, its disturbing and really there's no excuse, unless the original author can not be found.

I would not care if money wasn't involved, but clearly some bad guys are taking other ppl work and using it to make money. They think because its the internet they are free to break the laws.

But actually its illegal. I don't point fingers but at least I'm not trying to make money by stealing.

 

[mei2]

Just a quick update: Don't use downsampled Mono with Vrew. I've only checked the Korean language extractor, but if you make it mono and 16Hz you'll get roughly a third less text in the file. I tried a mono file at 48Hz and it worked almost as well as an unadjusted stereo track. Amplify it in parts and you might catch some more dialog but the program tends to interpret amped content differently. Also, I tried a compressed file (Threshold -16db, Noise Floor: -35 db) and didn't see any real improvement.
All attempts at removing all sounds but the dialog have resulted in sound with a changed character. Either muffled or 'glitchy' and clipped. There is background noise some of the time but the most difficult parts are when there's music that's as loud as the dialog.

Thank you. Very helpful !