akiba resident JAV subtitlers & subtitle talk★NOT A SUB REQUEST THREAD★

[Taako]

Well, I leave you some images of what was a test, so for now it is a new record, more than 2000 lines of text, very good, it is not perfect, there are some lines that the program repeats, but they are minimal, I hope you are back soon dark side and start creating more subtitles with this new system, I also leave the collab link.

Google Colab

colab.research.google.com

View attachment 3119823View attachment 3119824

What is the test? The movie you been working on?
2000 line is quite good.

 

[mrklhsnbrg]

Well, I leave you some images of what was a test, so for now it is a new record, more than 2000 lines of text, very good, it is not perfect, there are some lines that the program repeats, but they are minimal, I hope you are back soon dark side and start creating more subtitles with this new system, I also leave the collab link.

Google Colab

colab.research.google.com

View attachment 3119823View attachment 3119824

Hey I am also using Whisper. It is a very promising for Speech-to-Text model so far! I also tried it with Youtube Japanese contents. But with AVs, I happen to see the repeated sequences (mostly during AV's performers are having sex)

I wonder if you've had encounter this problem and if you have, do you have a solution for it? Thanks a lot

 

[soloporhoy666]

Hey I am also using Whisper. It is a very promising for Speech-to-Text model so far! I also tried it with Youtube Japanese contents. But with AVs, I happen to see the repeated sequences (mostly during AV's performers are having sex) View attachment 3120193

I wonder if you've had encounter this problem and if you have, do you have a solution for it? Thanks a lot

change to Whisper+VAD model, here is the link, this other Collab, it uses audio enhancement and repeated lines, it is easier to use and removes many of the repeated lines, repeated lines are a problem, but it is solved with this another option, leaving almost perfect lines of text, so far the best transcription option.

 

[quay2]

change to Whisper+VAD model, here is the link

Where?

 

[Taako]

Prinsipe said:
Hi Taako - are you a native japanese speaker? Like how can you tell if which is the right one, your sub or the whisper?

No. I would never call myself a native speaker. What I do have is some Japanese friends on other forum who sometimes help me with subs.

My skill in Japanese is very modest at best. I understand some things and speak very little. So I would NOT call myself a native. That would be insulting.

As I stated, my subs are usually 80-90% correct. Sometimes 70-80% . I would never place my subs any higher. Any though I'm confidence.

How do I know my sub for RCT-459 is correct? If you read what I wrote, I said "It's possible Whisper is right and my sub is wrong, but I doubt it.

With my first update, I compare my sub with what Whisper and Whisper simply did not compare to human translation. It left out lots of dialogue and made left sentence little to desire.

On my first Update; as I stated earlier, it did get some subs. It did. Why lie if it is functioning correctly? But... it also missed a lot of easy dialogue too. I won't lie about it. All of which is normal for software.
Not a big deal.

I knew this particular movie would be a good test because...
1. There's lots of talking.
2. More than 1 person talk as well.
3. There is background talking and some low talking.

Very ideal conditions to test Whisper imo. This movie isn't good but a good test nonetheless.

So on my second test, I wanted to watch the movie and sub together.
So I didn't compare it to my sub and went in blind.
1. To see how the movie and sub played well together.
2. Would I be able to understand the context of the movie given the sub.

My second test is rather uneven because I know what my sub say for each scene, kinda.
That said, I have to be honest, the Whisper sub did missed in letting me know the context of the movie.
It got some dialogue correct but the tone and context was lost.

If this was a new sub, I would not be able to tell you what the movie was about.
Instead I would have to look at the movie and guess the majority of what the movie was about.

Is this normal for Jav subbing? Yes I think so.

But when I sub, I try to sub every bit of dialogue and give you the full movie.
Yes it takes a long time to sub and yes it will never be 100% but I stand by my subs.

I tested whisper and showed a great result for me. I also run the same movie in capcut. Whisper and capcut shows the same transcriptions but there are lines and words that are better in whisper than capcut and vise versa. The transcriptions in whisper and capcut is similar. And from my observation, whisper construct sentences more than capcut. But my method is i am merging the two transcriptions to get a better transcription results by cleaning it up a little bit. From my experience of using different applications or software, whisper is by far the best and next to it is capcut. But that is just my opinion.

I will not disagree to anyone who results are good for them

Sorry if i disagree with your statement about "why accept something that is nothing and not accurate". Maybe you have a high standard, i get that.

It's not about high standards. It's about what you're willing to accept.
Everyone will accept what's necessary for them.

I choose to sub because machine translation were no use to me. And lots of the movies I like have no subs. Subbing my own just helps me understand the whole movie.

Because of whisper and capcut, I understand the story of the movie unlike before. So for me, whisper and capcut is not "nothing" at all, and about the accuracy - for me,

If you feel insulted about what I said you shouldn't. Again what I stated was my opinion.
As I stated, Whisper pick up a lot of the dialogue that even I missed. But... for this sub it miss a lot and the context wasn't there--for me.

Could it be the movie? Yes. I stated the movie was a good test because it has elements that make subbing somewhat difficult.
Different movies, different results. I believe that for any software being used.

Did Whisper pass the test for me? Again, it failed more than it pass for me...
But as a tool to help sub... absolutely it pass.

the best way to know if the transcriptions are accurate if you are a native nihongo speaker, or you studied nihongo to learn how to understand it, or if you know someone who are literate in nihongo.

That is a given for me and others who sub.

I don't believe anyone here; Imscully, George, Chuckie100, Darksider59, Mei2, Soloporhoy666, Makkdom, ding73ding, myself, and others believe our subs are 100% accurate. We don't.

In fact, if you notice we are very open and tell everyone we are NOT Japanese speakers and our subs are not 100% accurate. We also tell people if it's machine translations.

So, you need not to get upset. I think everyone here is very open and honest.
And everyone has a different opinions.

This is just my opinion. Peace and Happy Holidays everyone.

Thank you. Happy Holiday

 

[Prinsipe]

No. I would never call myself a native speaker. What I do have is some Japanese friends on other forum who sometimes help me with subs.

My skill in Japanese is very modest at best. I understand some things and speak very little. So I would NOT call myself a native. That would be insulting.

As I stated, my subs are usually 80-90% correct. Sometimes 70-80% . I would never place my subs any higher. Any though I'm confidence.

How do I know my sub for RCT-459 is correct? If you read what I wrote, I said "It's possible Whisper is right and my sub is wrong, but I doubt it.

With my first update, I compare my sub with what Whisper and Whisper simply did not compare to human translation. It left out lots of dialogue and made left sentence little to desire.

On my first Update; as I stated earlier, it did get some subs. It did. Why lie if it is functioning correctly? But... it also missed a lot of easy dialogue too. I won't lie about it. All of which is normal for software.
Not a big deal.

I knew this particular movie would be a good test because...
1. There's lots of talking.
2. More than 1 person talk as well.
3. There is background talking and some low talking.

Very ideal conditions to test Whisper imo. This movie isn't good but a good test nonetheless.

So on my second test, I wanted to watch the movie and sub together.
So I didn't compare it to my sub and went in blind.
1. To see how the movie and sub played well together.
2. Would I be able to understand the context of the movie given the sub.

My second test is rather uneven because I know what my sub say for each scene, kinda.
That said, I have to be honest, the Whisper sub did missed in letting me know the context of the movie.
It got some dialogue correct but the tone and context was lost.

If this was a new sub, I would not be able to tell you what the movie was about.
Instead I would have to look at the movie and guess the majority of what the movie was about.

Is this normal for Jav subbing? Yes I think so.

But when I sub, I try to sub every bit of dialogue and give you the full movie.
Yes it takes a long time to sub and yes it will never be 100% but I stand by my subs.

I will not disagree to anyone who results are good for them

It's not about high standards. It's about what you're willing to accept.
Everyone will accept what's necessary for them.

I choose to sub because machine translation were no use to me. And lots of the movies I like have no subs. Subbing my own just helps me understand the whole movie.

If you feel insulted about what I said you shouldn't. Again what I stated was my opinion.
As I stated, Whisper pick up a lot of the dialogue that even I missed. But... for this sub it miss a lot and the context wasn't there--for me.

Could it be the movie? Yes. I stated the movie was a good test because it has elements that make subbing somewhat difficult.
Different movies, different results. I believe that for any software being used.

Did Whisper pass the test for me? Again, it failed more than it pass for me...
But as a tool to help sub... absolutely it pass.

That is a given for me and others who sub.

I don't believe anyone here; Imscully, George, Chuckie100, Dark, Mei2, soloporhoy666, Makkdom, darksider, myself, and others believe our subs are 100% accurate. We don't.

In fact, if you notice we are very open and tell everyone we are NOT Japanese speakers and our subs are not 100% accurate. We also tell people if it's machine translations.

So, you need not to get upset. I think everyone here is very open and honest.
And everyone has a different opinions.

Thank you. Happy Holiday

Hi Taako -

Thank you for your response.

I am amazed how detailed you explain yourself. Sorry if I did offend you. And no, i am not insulted, it was just i also want to express my opinion on whisper and capcut. That is all. I am not upset, i just want to give credit for those who made whisper and capcut, for their hard work and specifically for giving it to us for FREE.

Your words are - "Why accept something that is nothing and not accurate?"

To give an adjective of "nothing" to the works of the programmers are an insult to them to be honest. For me, maybe, we should just choose the right words. Please don't get mad on this one. This is just my opinion too. Hehe.

You said - "How do I know my sub for RCT-459 is correct? If you read what I wrote, I said "It's possible Whisper is right and my sub is wrong, but I doubt it."

May i know what is your basis in doing your subtitles? Are you asking most of the spoken words in a movie to your japanese friends on another forum, or most of it was transcribed by you which you said that you are not a native japanese speaker and you said that you understand some things and speak very little. And also, how can you explain that the transcriptions in whisper and capcut are similar compared to yours. Please don't think that i am mad or something. I am just asking questions. Hehe. And also, please don't get mad at me. I believe that we are just expressing our opinions here and asking questions, okay? Peace.

Again, i am sorry if i did offend you or if i chose the wrong words.

Thank you for your feedback on whisper too. But I believe that the whisper and capcut transcriptions are accurate based on my test because the transcription results between the two different programs are similar in meaning. But of course not as accurate as human transcriptions and i agree with the bugs of whisper on repeated lines and some words are not transcribed well or missing, but for a program to do something like that is amazing for me.

Peace.

 

[Taako]

Hi Taako -

Thank you for your response.

I am amazed how detailed you explain yourself. Sorry if I did offend you. And no, i am not insulted, it was just i also want to express my opinion on whisper and capcut. That is all. I am not upset, i just want to give credit for those who made whisper and capcut, for their hard work and specifically for giving it to us for FREE.

This perfectly ok imo

Your words are - "Why accept something that is nothing and not accurate?"

To give an adjective of "nothing" to the works of the programmers are an insult to them to be honest. For me, maybe, we should just choose the right words. Please don't get mad on this one. This is just my opinion too. Hehe.

Again you're reading way too mcuh into this.
My opinion. Just because I look and accept things differently do no mean others have to.

For some reason you are taking things as an insult and I don't understand why???

You said - "How do I know my sub for RCT-459 is correct? If you read what I wrote, I said "It's possible Whisper is right and my sub is wrong, but I doubt it."

May i know what is your basis in doing your subtitles? Are you asking most of the spoken words in a movie to your japanese friends on another forum, or most of it was transcribed by you which you said that you are not a native japanese speaker and you said that you understand some things and speak very little. And also, how can you explain that the transcriptions in whisper and capcut are similar compared to yours. Please don't think that i am mad or something. I am just asking questions.

I explained most of everything, as precisely as I could. I mean... I don't know what you're asking???

Hehe. And also, please don't get mad at me. I believe that we are just expressing our opinions here and asking questions, okay? Peace.

Again, i am sorry if i did offend you or if i chose the wrong words.

Neither has happen.
You did not offend me. I am somewhat puzzled about what you called "insulting" the dev of the software...but
I'm not upset

Thank you for your feedback on whisper too. But I believe that the whisper and capcut transcriptions are accurate based on my test because the transcription results between the two different programs are similar in meaning.

This is good. As someone stated, Whisper is still "learning". So maybe it will get better and better as more people uses it.

But of course not as accurate as human transcriptions and i agree with the bugs of whisper on repeated lines and some words are not transcribed well or missing, but for a program to do something like that is amazing for me.

Peace.

Indeed, it's great AI has come so far.
As I stated, PyTranscriber, Whisper, and others are good tools for subbing. I still won't recommend capcut but I won't dismiss those who uses it.

Peace to you as well. It's nice to have a healthy discussion on a great forum as this.

 

[soloporhoy666]

Everyone is right, as far as I'm concerned, Whisper isn't perfect, but it works for me, I've been creating the subtitles in Whisper and finally I could understand the plot, I don't mind that sometimes I changed some words (sucks my hand) it is obvious that it refers to something else hehehe but I can survive with those failures, just like the subtitles that I find and change the meaning from mother-in-law to mom or daughter, to give a different meaning, if the human makes changes, why AI can't? well to finish Whisper is much better than the other similar options, when translating into my language it is quite understandable, the grammar is not perfect but it works for me, that is why I am excited to use this AI, it gives me results than with other options ever I managed, obviously it will never be perfect as a human translation, where the interpretation and the emotion of the person is essential, (let's take let's go I feel like making love to you) just an example. have a nice party.

 

[Taako]

Everyone is right, as far as I'm concerned, Whisper isn't perfect, but it works for me, I've been creating the subtitles in Whisper and finally I could understand the plot, I don't mind that sometimes I changed some words (sucks my hand) it is obvious that it refers to something else hehehe but I can survive with those failures, just like the subtitles that I find and change the meaning from mother-in-law to mom or daughter, to give a different meaning, if the human makes changes, why AI can't? well to finish Whisper is much better than the other similar options, when translating into my language it is quite understandable, the grammar is not perfect but it works for me, that is why I am excited to use this AI, it gives me results than with other options ever I managed, obviously it will never be perfect as a human translation, where the interpretation and the emotion of the person is essential, (let's take let's go I feel like making love to you) just an example. have a nice party.

Agree. And as I stated it got a lot of the dialogue somewhat right.
In the case of RCT-459 i try to approach it the 2nd time as someone not having a previous sub.
Although, it didn't help me undersand the movie, it did help fill in some words I missed.

We all have to be honest in our assessment of any software, that's the only way to find the problems and resolve it.

And also let's not forget, different movies will yield different results.
Whisper, like many other softwares will have bad to good results with different movies. And be even more useful, except capcut lol


Anyways, I am happy there's another software I can use. And I am never afraid to ask for help as I have stated many many times I'm certain I'll install Whisper in the future but not capcut lol

 

[Prinsipe]

Below is a sample of capcut transcription. Seems decent to me, and with that result, I understood the movie’s story. I did not edit anything or clean up anything on that. It is not that accurate, yes, but the overall context of the story is there. So, for me, capcut is still a good application to generate transcription.

 

[xsf27]

Hi there, I've been using Whisper AI to try to translate some JAVs and am using VLC player to rip the audio from the video files in order to upload.

I apologise if this has been gone over previously, but I was just wondering if there are any settings that I should be activating on VLC prior to ripping to ensure maximum legibility for Whisper? Or is there another better software or website that is better suited for the task?

 

[Taako]

For RCTD-459, I just checked the first minute (36 seconds to 1min) , and of course as it's AI, it's wrong.
..............

00:00:36
-Here it is.
-Thank you

00:00:38
-Put this on (the table)
-Okay

00:00:44
-Oh ? There is a yogurt ...
- Try to be careful

00:00:46
Thanks. It makes me happy.

00:00:48
-It's the morning and I'm already pumped
- Really ?

00:00:50
It's the same thing everyday

00:00:51
Hello

............

Yes.
When I compare it with my sub. I stated it wrong right out the gate lol

I think this software is best used as a base, as myself and others suggested.
It can provide some timecodes and some subs...but for real accuracy...not so much.

 

[maload]

i found something on the net that very funny
its about how to use "mango "
its not just " pussy "
it can be something else too
and i guess lots of movies have these word in the movies for sure
so its useful

manko マンコ, omanko おまんこ (Vulgar, Sex-Related Term)

What is manko? Meaning in Japanese.

www.japanesewithanime.com

 

[Taako]

I decide to start posting on this thread because it's not for posting subs, I think we are forgetting what that thread is about.
Anyways if a mod, wants to take a look and appropriately start cleaning it up, that would be nice

So I'm going for dinner/drinks soon and decided to test Whisper for a couple of hours.
An hour movie, mp3 of course. The movie I know only has "talking" for the first scene.
I want to know...
1. Can Whisper find the subs?
2. Are the subs accurate?
3. Will it make up subs, for a movie where subs don't exist

Anywayyyyys, Happy New Years

 

[Prinsipe]

I

I decide to start posting on this thread because it's not for posting subs, I think we are forgetting what that thread is about.
Anyways if a mod, wants to take a look and appropriately start cleaning it up, that would be nice

So I'm going for dinner/drinks soon and decided to test Whisper for a couple of hours.
An hour movie, mp3 of course. The movie I know only has "talking" for the first scene.
I want to know...
1. Can Whisper find the subs?
2. Are the subs accurate?
3. Will it make up subs, for a movie where subs don't exist

Anywayyyyys, Happy New Years

I thought you already tested whisper and you have detailed explanation about the results based on your experience on other thread? And i thought those questions have already been answered by you.

 

[Taako]

I

I thought you already tested whisper and you have detailed explanation about the results based on your experience on other thread? And i thought those questions have already been answered by you.

Nope I never tested whisper. I did compare it to my subs.
This was a different experiment.
But since you asked.
the results seem to been near the same, if not worse.

1. It captured some subs.

2. missed some verrrrry easy subs with no background or cross talking involved.

3. And it made up words that wasn't within context given the scene. Example: the sub say Good night. but the actress is saying dozo.

4. It had numerous errors of verbs and etc. I hate verbs. I hated it in school and I hate that I know when it's being misused.

5. This is weird. It seemed to miss the tone from 2 of the 3 characters...
but got part of the tone from the other character...2X, in a short exchange of dialogue.

6. Another strange thing. Of the dialogues it picked up, If it had low talking, it was wrong.
However, in some cases it tried to changed the sentence structure to makes it work. Essentially it tried to change 3 sentences to make the dialogue work as a whole. The change was noticeable because I know the sub.
But...if you don't know Japanese. You might think it's correct. wtf lol

7. It still missed lots of dialogue for such a short movie imo.

8. And it took way longer than I was expecting.

9. This sub was easy because even without the sub, you can guess what's going on in the movie. Again no background and cross talking. If anything some low talking from the man.
I don't blame Whisper for all of that. It can't be helped. But it should be noted anyways.

10. I didn't increase the volume or anything. So the mp3 was as-is.

My opinion of Whisper hasn't change. Use it as a base for your timecodes and whatever subs it gives you.... But make sure the subs are correct because 80%-90% of the time they are not.

Happy 2023

 

[Prinsipe]

I disagree of the ''80-90 percent of the time they are not".

So it means that you gave a feedback about whisper before without even testing it by yourself? Lol.

I assume you used the large model in your test.

Anyway, that is your opinion. Thank you for your feedback.

Happy 2023 too.

 

[mei2]

....

My opinion of Whisper hasn't change. Use it as a base for your timecodes and whatever subs it gives you.... But make sure the subs are correct because 80%-90% of the time they are not.

Happy 2023

Happy 2023 @Taako. May I ask which movie/scene was it that you tested? I am working to calibrate Whipsers parameters and would love to test my test settings on the same movie/scene for benchmark. That would be great!

 

[Taako]

I disagree of the ''80-90 percent of the time they are not".

You are allow to disagree. I didn't realize my comments and experience would trigger some people lol

So it means that you gave a feedback about whisper before without even testing it by yourself? Lol.


I assume you used the large model in your test.

You would be wrong on both accounts as it clear you never read my first time talking about Whisper and a particular sub

And now you assume, I haven't read any of @SamKook experiences/findings/suggestions about Whisper, amongst his many other suggestions about various tech issues.

Not to mention reading and getting help from others on the platform lol

Anyway, that is your opinion. Thank you for your feedback.

Happy 2023 too.

Omg my opinion. I didn't know that. Said no one ever lol

 

[Taako]

Happy 2023 @Taako. May I ask which movie/scene was it that you tested? I am working to calibrate Whipsers parameters and would love to test my test settings on the same movie/scene for benchmark. That would be great!

Happy 2023
The movie is called hmm-004
and Confessionale