Search results

I've been documenting some more of the plot connections for the Slave Color and spinoffs for this thread and been especially trying to document the films that have connections. More often than not those connections come from the female dominants. Most of the time they just taunt and trash...

My best luck so far has been no_speech_threshold at 0.6, log at -1.0, and condition_on_previous_text as False. Beam size something like 8 or 10. This is a good general use for recent HD videos. I imagine that lower video quality could use other solutions. From there, open it up in...

Thanks to Whisper, there will probably be English subtitles for all of these within the next few months. I may put a few of my own edits on SubtitleCat, but I assume there's going to be more translations off of Chinese translations via Whisper. It has been 8 months since the last entry (a...

False for "condition on previous text" lets Whisper start doing things like subtitling (crying) and (grunting), but I also get (speaking in foreign language). No shit, Whisper? That is what I am here for! That is some of the value in running Whisper multiple times per project. One of them...

If you are using the command line (which I assume is what you are doing based on the "--", it is "--condition_on_previous_text False". You do not do any equal sign on the command line.

I think this thread will be a good place to debug subtitle files, but it would be a lot easier if we were allowed to upload .vtt and or .srt files. I can't imagine there being a reason why not. Otherwise, standard practice should be to rename them to .txt so they can be opened and viewed...

As far as I can tell, you can't just run things with the simple command line after that. You need to start worrying about the different parts of the process. Not sure if that has changed or if it will change, but the documentation does not describe the process well. There is logit biasing or...

Has somebody done that though? Once you have a fine tuned model, Whisper is vastly more difficult to run as far as I can tell.

Damn, that probably is a result of the large-v2. I don't really understand the interplay between No Speech and Logprob. I think hallucination may end up solved via Logprob but I don't know what values to even guess at for it.

No Speech at 0.08 is very low. Do you get lots of phantom speech?

If you're trying to get away with just Small, and it only takes a few minutes, you can try to do runs with a low temperature (but not zero) and like best_of 10 or something. Part of the key to Whisper is doing settings that try to compensate for whatever your computational stuff is limited by.

I have a frequent problem of the first 30 seconds not having translations or having bad translations, and then the chunk after the 30 second mark to be rushed and horribly mistimed.

I made a thread for Whisper setup, parameters, and discussion https://www.akiba-online.com/threads/whisper-openai-automatic-english-subtitles-for-any-film-in-any-language-an-intro-guide-to-subtitling-jav.2115103/

OpenAI made massive waves in 2022 when they released Dall-E 2 and ChatGPT, but that year they also released Whisper for Automatic Speech Recognition, meaning transcripts. Not just transcripts, but direct-to-English transcripts from any of 97 different languages, including Japanese. In...

Set "condition_on_previous_text" or whatever it is to False. In theory, it is there to make it learn into consistent style and slightly bias the results towards specific topics that the entire source is about. In practice specifically for porn films of languages other than English, there's...

Do some parameters get better timings? There's the known issue with not getting good timing within the first 30 second window, but there's some pretty bad mismatches all through some transcripts while others are great.

DeepL is theoretically better, but there's probably some value in doing direct-to-English with the same deep learning model rather than taking the transcribed output and feeding into a second deep learning model that isn't specifically designed to work interact with the first. There's just an...

I looked at the help, and tried all combintaions of "" (), and []. Searching for help with Whisper online is just extremely difficult because it is new, the name sucks, and because virtually nobody seems to be using the command line. https://blog.deepgram.com/exploring-whisper/ has examples...

What is the syntax for Temperature on the command line? in Python it is temperature=(0.2, 0.4, 0.5) and best_of=3 to use those three different temps, but I can't get a specific spread like that to work on the command line.

Thank you, I had seen the Temperature discussion but not that optimization blog.

For some reason many subs generated via Whisper will have a bunch of subtitles start exactly at 30 seconds, and totally rushed through and offset. Most of the time the timing is great, but I've seen this initial rush a few times and I can't tell how/why it is happening like this. For the most...

It is worth messing around with --suppress_tokens feature to see if this can be adjusted. Argument of "food" or ["food"] (not sure when to use which) will make it avoid using the word food, or words that are something similar.

I totally missed that Slave Color Married Woman RBD-468 had a really good English subtitle. I don't know if I had missed out on the film the first time around, because wow does this feel like one of the better entries in the series. They trash talk her like hell, she puts out a lot of crying...

I think Whisper Transcribe + Google Translate is a sidegrade. Not seeing consistent advance to it. Some lines better, some are worse. An adjusted Whisper model will likely beat both.

I am getting close to the point where I could make a Whisper thread that discusses how to best use it. I want to see if I can get one of the JP focused models working though. "--conditon_on_previous_text False" will prevent some of the repeated lines, but lead to worse translations everywhere...

It depends what you're going for. The value of Whisper isn't to get the highest quality translation possible, it is about lowering the effort bar to get something good enough. Medium Whisper, with well-chosen parameters and some manual editing from just watching the move and editing as you...

Seriously, the gap in quality between Whisper and most JAV subtitles is huge. RBD-179's translation on Subtitle Cat's first 2 minutes: Now let's look at even the Medium basic model from Whisper. The only thing it gets wrong is who married Mr. Kazuo. Her mother married Kazuo. I see that...

For Whisper, there's probably better settings for "compression_ratio_threshold" that are more suitable for dialogue during sex scenes. There is a new "initial_prompt" option that might be able to kick it into some dirtier talk, but there's not really a lot of guides that are useful, and movies...

Good handmade subtitles will beat Whisper for JAV, but does that type of good subtitles even exist? Even the paid ones that are seemingly done by hand miss a ton of lines. Whisper gets wildly better coverage of what the men say during scenes in Attackers movies than any subtitle I've seen...

The problem is getting a good dataset. Whisper is trained on 600k hours of "quality" transcriptions, broken down into 30 second pairs of audio and transcript, and aggressively filtering out transcripts that were machine generated. There's not going to be a good set of "porn with hand...

I've been impressed by Whisper. I am running it locally, which I can handle the Medium model but not the large. Various thoughts on it and how it relates to JAV. I think it is worth making a different thread for discussion of optimization, as there are many parameters. Whisper is best at...

Subtitle Cat has the few HUNBL subtitles that exist. Seems like all of them are just based on Chinese subtitles, but the plot and dialogue are simple enough for the translations to work.

I think that is Hunter Black's best film to date. There's just a lot more actual talking and emotion than the vast majority of what they put out, the bulk of which is more like a MAD film than an Attackers film. There's a lot of good stuff in their Immediately Irrumatio series (a...

Thank you, I saw that thread but it seemed like Whisper is really just piggybacking on that thread. Whisper optimization should probably be its own thread but for now I'll post there.

Whisper seems really impressive. The Large model is a lot to ask and very few people will have a PC capable of it, but the Small model is pretty reasonable and gets pretty good results that can be edited. Medium better still, and there's a few adapted models that claim to have better...

SSPD-006 would probably be super well regarded if there were better rips out there. There is a DVD version but the rip that is up there is really low quality. What is the plot? It seems like three totally unrelated r*** scenarios for the women, and then they are all brought together as...

https://attackers.net/works/list/series/2396 I imagine this official list is wildly incomplete because there's been so many. It is one of Attackers' longest running series. It is a mix of slavery slavery, and just blackmail where the woman is allowed to go free, or just briefly.

Is there a go-to program for extracting Chinese subs that are hardcoded? Lots of good rips have Chinest subs now. Slave Investigator 1 has good Chinese subtitles on a newer HD rip that I want to translate. Or even just a video player that lets me press a button to add a subtitle to an SRT...

Are the R18 Player videos you can download going to be totally worthless once R18 fully goes offline, or are they going to be able to be run in a media player after the validation system is gone?

In this case you can just look to count through 1, 2, 3, 4, 5, 6, etc. I thought there were gaps but the search was just so out of order that I missed a few. I don't think places in Akihabara would have VHS tapes now, and if there's some random DVD I wouldn't trust it to be something better...

Getting better quality rips comes down something like "The Chinese will save us!" If there is a higher quality rip that appears for something from the late 00s, more often than not it seems to have hard-coded Chinese subtitles. There is a new and growing market there that is big enough to pay...

The best coverage of old films of Attackers looks like here, at jav.land, searching by the label. They retroactively started calling the RB films ARB, which I guess would be the pre-RBD era. https://jav.land/en/label/o3my63.html?page=44 https://jav.land/en/label/l30rp3.html?page=42 SHKD...

For those that don't know the history, the D in Attackers film codes stands for DVD. So, for example (not sure if I am getting the era right), there is both an RB-040 and an RBD-040, with RB being the VHS version and RBD being the DVD version. People just kind of treat all of the old films as...

R18 is done for good, DMM blocks US IP addresses and accounts. Attackers themselves have spotty coverage that far back, so it is even hard to get any information. JavLibrary isn't complete and doesn't have the plot summaries that are on the sale sites. I get that it is super weird to be fans...

Ah, that set of wallpapers reminds me how great SHKD-184 is. Shame there's no decent rip out there. SSPD-060 looks mislabeled? Which code does that actually belong to? EDIT: Oh, just SSPD-006. Many of those old SSPD films deserve better preservation than they got. Even the versions you...

The first Slave Investigator, ATID-186, has better rips than before, and has some good Chinese subtitles, but not English translations of them. Is there a standard process for ripping the subs, or a good archive of sub files that can be translated? SSPD-035 is great. That entire series is...

SHKD-292 is part of this collection, if anyone can't find the original rips. I think just the most extreme gangbang scene is what gets included in these collections-- they aren't the Full Preservation editions that the Slave Color or Slave Island got (which are were basically 4 packs or 6 packs...

Until seeing taking the time to look through Attackers' site, I had no idea that the New Slave Investigator series was the actualy series as the old Slave Investigator series. As in, it is a continuation of the same numbering. New Slave Island was a reboot, but New Slave Investigator is...

That would basically double the amount of work in the post for me, and it has a lot of shortcomings. The Series pages aren't complete, because of the title changes and because the first entry usually isn't properly tagged. There's a fair number of entries that have been scrubbed from their...

Just grab the HD ones with the most completed downloads. While it may show zero streamers they are more than likely not "dead". People are sharing the actual movie but not all of the junk files, or the true streamers are only online part of the day.