Whisper and its many forms

[DScott]

I use media player classic home cinema: https://github.com/clsid2/mpc-hc/releases
It's open source so no crappy ads or bloat.

Just did a quick test and it has no issues with subtitles and ts files. It wasn't made from it though, I just used a renamed random sub so I don't know if that matters.

I found an interesting player called mpv player it is open source. It has a very very clean interface, reviews are good, and image quality better at hi-res. My only issue, never satisfied I guess, is that it has absolutely no play opttions beyond play and stop. I don't know if you've every tinkered with VLC but what a program! So many great features! This mpv player has none. And i mean NONE. Buuuuut playback is pristine and it played .ts/srt combo. My only real concern is that if I am uling that combo the enduser will have to have something to play that. Iknow that VLC is wildly popular so that arrangement wouldn't work. I did a lot of reading and with regard to VLC fixing that issue there is a myriad of the ol 'it's the other guy's fault' If it were just for me I would just continue to use VLC and use mpv for isolated ts/srt combos however with the idea of uling to Akiba I think that I'll continue to convert to mp4 first so that the srt will play in vlc, mediaplayer, etc. Unless, of course, someone suggests a better option. Thanks, as always, Sam..

 

[SamKook]

MPV has tons of options, but basically no user interface, it's all keyboard shortcuts. I also use it since it has a plugin to view VR videos in 2D.
I pretty much only use VLC to play bluray, it's the best I've found at that.

Like I said earlier, ts is a pretty bad container so it's a good idea to use another instead. Just make sure you don't re-encode the video and that you simply remux it(meaning taking the video and audio from inside the ts file and putting it back into another container). It should be very quick, seconds to a minute or 2 at most.

 

[DScott]

MPV has tons of options, but basically no user interface, it's all keyboard shortcuts. I also use it since it has a plugin to view VR videos in 2D.
I pretty much only use VLC to play bluray, it's the best I've found at that.

Like I said earlier, ts is a pretty bad container so it's a good idea to use another instead. Just make sure you don't re-encode the video and that you simply remux it(meaning taking the video and audio from inside the ts file and putting it back into another container). It should be very quick, seconds to a minute or 2 at most.

Yeah, thanks Sam. ThaT view VR in 2d is very attractive. What's the name of the plugin? I have never liked .ts but it seems like the most used by sites delivering free JAV titles like missav.com etc. I'll keep converting to mp4 I think but I believe that I'll explore mpv a little more. Thanks again Sam..

 

[SamKook]

https://www.akiba-online.com/threads/how-to-watch-vr-videos-with-a-pc-player.2123024/post-4665335

TS is a broadcast container mainly. Good for that but not so good for local playback. Usually the software you use to download those will convert directly to an mp4 but I guess yours doesn't.

 

[Walle12]

VR-to-flat is like going from video to pornographic magazines... A step backwards.

Mates, what you need is a Quest 2, you can get them for less than $150 second-hand....

 

[SamKook]

I thought about it but the meta login requirement is a dealbreaker for me so it gets considerably more expensive. I will one day buy a VR headset but now is not the time.

 

[Walle12]

I understand you... I don't have a Quest 2 either, nor will I ever buy a Meta headset...

 

[Fire117]

How do I get audio from a video in the first step to using whisper?

 

[SamKook]

How do I get audio from a video in the first step to using whisper?

Check the "Extracting audio" section of the whisper tutorial: https://www.akiba-online.com/threads/whisper-and-its-many-forms.2142559/post-4719912
Just click "Spoiler: Extracting audio" to show it.

The "Remuxing to an audio only container" subcategory is the easiest way to do it.

 

[Fire117]

That's done, what do I do next?

 

[SamKook]

Follow the instruction for the online version by clicking the next spoiler tag: "Spoiler: Colab Whisper+VAD"

 

[Fire117]

Follow the instruction for the online version by clicking the next spoiler tag: "Spoiler: Colab Whisper+VAD"

The Instructions are confusing, what do I do exactly.

 

[SamKook]

Ask in that thread about what confuses you exactly. I'd just repeat the same thing written there to explain how to do it unless I know which parts are confusing since it's clear to me.

 

[Fire117]

Ask in that thread about what confuses you exactly. I'd just repeat the same thing written there to explain how to do it unless I know which parts are confusing since it's clear to me.

Mount drive failed, so what do I do next?

 

[SamKook]

Make sure you use https://colab.research.google.com/g...AV/blob/main/notebook/WhisperWithVAD_mr.ipynb
and that you've signed in to your google account, followed what's inside "Spoiler: Google Drive Upload" and gave the appropriate permission to give access to it.

Alternatively, don't mount the google drive(skip that option) and just do what's in "Spoiler: Session Upload" instead.

 

[Fire117]

Make sure you use https://colab.research.google.com/g...AV/blob/main/notebook/WhisperWithVAD_mr.ipynb
and that you've signed in to your google account, followed what's inside "Spoiler: Google Drive Upload" and gave the appropriate permission to give access to it.

Alternatively, don't mount the google drive(skip that option) and just do what's in "Spoiler: Session Upload" instead.

Damn WIFI is going red on me, been a constant problem for the past week. Have to do it all over again.

 

[Fire117]

Finally got it done, but the SRT file has inaccuracies. Plus dialogue said and no subtitles in English for some of it when I play it with the video file.

 

[SamKook]

Most people here don't make accurate subtitles, they simply take a machine translation and they take the time to edit it to fix the obvious mistakes, which often means interpreting the scene since they don't understand japanese, very few actually do to some degree.

Whisper does a decent job as a base and is very simply to use so if you want to give it a shot yourself(there's way more request than subs actually getting made here so don't insist too much, just ask once), you can check the tutorial in my signature and ask around if you want to tweak things further than the basics.

Then you use a software like subtitle edit or aegisub and edit the sub until it's good enough for you.

Like I said in that other thread, it's a decent base, but that's all it is, a base. To make it better, you have to work on it, which is what people who share subs on here do. No machine translation will give you a perfect result, technology isn't that good yet. Whisper is just the best option around that's very easy to use.

For the parts where there's nothing, you can use audacity(or any audio editing software) to try and enhance the audio to make whisper recognize it or try to sound out what is being said and see if you can match that to actual romaji.
There's a list of common words shared in the big subtitle thread if I remember right which can help.

You can also try to tweak the settings following peoples suggestions to see if more can get detected and merge or correct into 1 new version by comparing the 2 or multiple versions you got.
There's some drawback to making whisper detect more stuff which is why it's not the default but it can be good for specific parts.
The 3rd version of the colab ( https://colab.research.google.com/g...V/blob/main/notebook/WhisperWithVAD_pro.ipynb ) allows for more in depth tweaking of the settings, but just changing the "model_size:" to different version of large(v1 to v3) in the link I said to first use will give you different results.

 

[Fire117]

I wouldn't be able to make it better myself. Only foreign language I know a bit of aside from English is a little bit of Spanish.

 

[SamKook]

When you don't know, you try to guess. As long as it makes sense with the story, you won't notice whether it's accurate or not.

Just look at the disclaimer Chuckie100 adds to the subs he shares:

I used Whisper Colab to create this Sub 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's happening in the scene. Anyway, enjoy and let me know what you think.​

To get a completely accurate translation, you'd have to pay hundreds of dollars since it's very hard work to do completely right.