software translating subtitles

[Jan de Uitvreter]

Hi, I'm the developer of AI.opensubtitles.com.
It is an online platform that allows users to transcribe and translate their audio....

Hello dear 'ai.opensubtitles' salesman.
The double and free build-in speech recognition systems in SE seems fine to me.
Both are also rapidly extracting the audio from the video before the engines starts.
My question as end user: Why should I pay for your AI.opensubtitles?

SE engines info:

SE got two different integrated speech recognition systems, respectively: Wisper and Vosk/Kaldi.
Both are working just fine and up to date (even today I had a update)

You can choose from several Whisper models like: tiny 74MB until
large-v3 (3.1 GiB) and many more other sizes in-between
Whisper engines are:
Open AI
Purfview's Faster-Wisper
CCP
CCP cuBLAS
Const.me
CTranslate2
Stable-ts
WhisperX


A view Vosk/Kaldi models:
vosk-model-en-us-0.22-lgraph
vosk-model-language XXXXXX-0.6-lgraph

Vosk/Kaldi model I'm using: Dutch (large 860 Mb)
But there are much more other language models to choose from,
like for example: English (medium 128Mb) or English (very large 1,8GB)
or Chinese, Arabic, Russian, Filipino, Swahili etc. etc. etc.

I'm pretty happy with these build-in engines and they are 99% accurate onto Dutch language
@ 1 hour transcribing I had only 1 error onto a Dutch woman's name: 'Truus' (Engine result was Trus)

Kind regards, Jan de Uitvreter

EDIT
Note:
I'm only speech transcribing Dutch to Dutch
from other languages to Dutch makes not much sense for us Dutchies

[ai.opensubtitles.com]

Hi Jan, Oude Kaaskop,

You are free to choose whatever service suits your needs. No hard feelings.
But I have to say, you are missing out on the best AI.opensubtitles has to offer. Whisper models are not very reliable, as you will soon find out when you try to transcribe or translate anything other than Dutch. And Vosk/Kaldi models are not much better, either. They are outdated and limited in their capabilities.
On the other hand, AI.opensubtitles uses the latest and most advanced AI models from various providers, such as AWS, Assembly AI and DeepL. We have a wide range of language models to choose from, and we constantly update and improve our site.
Our service is not cheap, but it is worth every penny. You get what you pay for, and more. You can try it and see for yourself. We are confident that you will be impressed by the quality and speed of our service, and that you will join our growing community of satisfied customers.
But hey, if you are happy with your SE engines and their 99% accuracy (which is probably an overstatement), then good for you.

Kind regards,
The AI.opensubtitles.com salesman

https://ai.opensubtitles.com

[Jan de Uitvreter]

Yup, trotse oude kaaskop hier, OS lid sinds jaar van oprichting

It's absolute not an overstatement you should try yourself.
I was really amazed about SE's Dutch to Dutch speech transcription accuracy myself.
Vosk/Kaldi did the job fast and right, it not feels out of date.
For me, Speech transcribing was a onetime project. There's no need for me to us often.
But right now SE's integrated models seems to be alive and kicking and does the job.

Cheers, Jan de Uitvreter

[ai.opensubtitles.com]

Hi Jan,

I appreciate your loyalty to SE's integrated models, and I'm genuinely thrilled that you found a solution that works for your one-time project. It's heartwarming to see someone advocate for the 'free and fast' options out there. After all, who needs reliability and quality when you can have a swift 99% accuracy, right?

I understand your enthusiasm for the budget-friendly alternatives. It's almost poetic, like a community-driven effort to democratize AI for the masses. Your commitment to showcasing that even with limited financial means, one can still enjoy the thrill of transcribing foreign movies is commendable. I can picture the heroics now: 'Saving the Day, One Whisper Model at a Time.'

But let's talk facts. We don't include those 'free' models on https://ai.opensubtitles.com/ because, well, we are in the business of delivering premium quality. Why settle for mediocrity when you can bask in the glory of cutting-edge technology? We aim to provide a service that sets the bar high, and if that means charging a bit, it's because excellence comes at a cost.

Your dream of a completely free version has some merit. We're actually considering it, partly using the same AI models that fuel the 'freebie software' you're championing. An online subtitle editor rendering those locally installed relics obsolete, ushering in a new era of accessibility. Perhaps then, even the most ardent fans of 'free and fast' will see the light.

Interestingly, despite the allure of SE, many of your fellow countrymen have found solace in https://ai.opensubtitles.com/. Our platform, with its refined selection of top-notch AI models, has become a haven for those seeking unparalleled accuracy in Dutch transcription and translation. It seems the allure of excellence transcends borders, even those guarded by SE loyalists.



Cheers to diversity in choices, Jan. May your transcriptions always be swift and your errors be few.

Kind regards,
The Unwavering AI.opensubtitles.com Dev

https://ai.opensubtitles.com/

[SmallBrother]

First of all: best wishes for everyone for the new year!

@Oss

It is always big discussion, there are users who saying even not perfect subtitles are better than none. And then there is other users and views on this point.

We try to be neutral.

Please note something very important what subtitler hendrikeduard1 said:
I feel like I'm becoming redundant due to the promotion of AI.
(...)
If AI is going to be used from now, I'll finish what I'm working on and stop doing it....

This is not the first time nor the second time I heard this from various subtitlers.
Reality is, and more and more will be, that by being 'neutral' as you say, the actual effect is not neutral at all. We will end up with mediocre socalled 'retail' subtitles fastly and cheaply created by all the streaming services and what is missing, will be filled by even more mediocre machine generated subtitles. And please, let's face it: those machine translations are not just "not perfect". They are pretty far from being perfect.

thats why we REQUIRE any uploads which are AI generated to mark it so.

If admin allows that.

And here is another problem.

During the past 10 years of me being admin on OpenSubtitles, I have spent quite some time and energy on trying to create a subtitling community, especially in the Dutch section. From protecting credits and community peace to coaching users with making subtitles, either on a one-to-one level, or by extensively writing forum topics. And motivating others to do so as well. Although nobody, including myself, is perfect, I think I did pretty okay.

But lately I am spending most of my admin time on spotting and deleting machine translations and often having the same endless discussions about it. The latest argument in such discussions is:
- Why did you delete my machine translation, while OpenSubtitles is promoting it left and right, even asking money for it? It doesn't make sense!

I must agree: it doesn't make sense. And it's frustrating to be mopping the kitchen floor and trying to get it dry, while others are throwing buckets of water non-stop. And that frustration is getting greater every day, especially with the site and the forum being flooded (or, may I say "spammed"?) with the promotion of all sorts of machine generated subtitles.


@ai.opensubtitles.com

"premium quality, the glory of cutting-edge technology, excellence, refined selection of top-notch AI models, unparalleled accuracy, symphony of precision, speed, and versatility, the Michelin-star chefs of the transcription world, meticulously trained and refined for linguistic perfection, finesse, accuracy, and a touch of AI elegance, a gourmet menu of cutting-edge AI models, a transcendental experience in language processing"...
Well, I must say that I admire your poetic skills, but to me this sounds more like a salesman selling his own goods, rather than an independent review of an end-user

Don't get me wrong, I respect your efforts and trust your developing skills. I don't doubt the accuracy of the transcripts or translations done by ai.opensubtitles.com. In fact, that's exactly where the problem is. For English subtitles of an English spoken movie or tv-show, a speech recognition transcript may be excellent as subtitle, as they should (must) be 100% verbatim. For other languages this is not the case at all. Subtitles must abide by a large set of guidelines. An accurate translation is actually only a 'minor detail' and typically very much in the way of other guidelines. So a 100% accurate speech recognition which is 100% accurately translated into another language will actually result in poor subtitles.

Please see the guidelines for Dutch subtitles as described here: viewtopic.php?f=38&t=12639 (I suppose the Dutch language is no problem) and you will have an idea what it takes to make a proper subtitle. And most probably you will understand why any kind of machine translation is very likely to fail.

Further a little question mark on your statistics:



With more than 7 million OpenSubtitles users, I would say 7000 AI users is not really impressive. Besides, it doesn't indicate how satisfied they are with the service. From my experience, some users are very impatient about their favorite tv-show and their standards are not very high, that may be it.

That being said, soon I will try out your AI service and I will sincerely do my best to prove my wrong. I will let you know on viewtopic.php?f=1&t=1969&start=90#p50132

[ai.opensubtitles.com]

Further a little question mark on your statistics:



With more than 7 million OpenSubtitles users, I would say 7000 AI users is not really impressive. Besides, it doesn't indicate how satisfied they are with the service. From my experience, some users are very impatient about their favorite tv-show and their standards are not very high, that may be it.

That being said, soon I will try out your AI service and I will sincerely do my best to prove my wrong. I will let you know on viewtopic.php?f=1&t=1969&start=90#p50132

I share your arguments about the quality differences of transcription and especially translation ( and especially from English) using AI models. But even this might be overcome in future, because if you can train an human to do it, you can train an AI too. But 100% accurate subtitles won't be something very high on the priority list of GTP developers / investors right now.

Because of this, I won't publish the 'upload to OS.org' feature I worked on a month ago, so only download for an translation is possible now, but no re-upload of the translation . The site owner told me about the community admins reservation about AI content, and I can really understand your point of keeping the quality standard.

The poetic description of the transcription was more a reaction to these annoying "Why should I pay for anything" comments from the SE salesman, than a realistic review.

I'm looking forward to your review of https://ai.opensubtitles.com.

About the stats ... I wish I had 7000 paying customer, these are just visitors, just like the 7 million on OS. But it's still true that I make most money with customers from land of good cheese and weed.

[SmallBrother]

Because of this, I won't publish the 'upload to OS.org' feature I worked on a month ago

Oh my god. Please don't.

The poetic description of the transcription was more a reaction to these annoying "Why should I pay for anything" comments from the SE salesman, than a realistic review.

Advocate of the devil: His avatar may give the wrong impression, but Jan de Uitvreter is NOT a salesman of SE (even SE being free). Nikse from Denmark is the developer of SE (see www.nikse.dk ). Jan de Uitvreter is just a satisfied Subtitle Edit user. And an excellent subtitler, btw.

I don't think Jan de Utvreter (or anybody) questions the fact that sometimes we have to pay for things. I think, after his findings with the speech to text from SE, his question was legitimate. After all, only ONE error ("Trus" vs. "Truus") IS actually (more than) 99% accuracy.

I'm looking forward to your review of https://ai.opensubtitles.com.

I just posted it: viewtopic.php?f=1&t=1969&start=90#p50183

[Mazrim Taim]

(....)

I share your arguments about the quality differences of transcription and especially translation ( and especially from English) using AI models. But even this might be overcome in future, because if you can train a human to do it, you can train an AI too.
(....)

I highly doubt that assumption. As someone once said: "Language isn't a formal system. It is a glorious chaos".
That's the difference between humans and machines. Machines don't do chaos, they follow rules. We humans don't, we are more flexible in that respect.

Understanding of language is not a series of complex computations, nor is it the consulting of an ever expanding database.
Humans use an intricate system of associations between lexical concepts and their context. Machines don't. They are, as yet, unable to grasp context.

Simply put: they do as they are told. However, they do not see. Nor do they understand.
Given enough time and effort this might change, but we will need a sentient A.I. in order to achieve this.

Until the science behind A.I. stops depending solely on machine learning (essentially programming) and starts trying to mimic the evolutionary consciousness that has developed in human beings over millennia, it will not so much as come close to creating true artificial intelligence.



For now, A.I. assisted translation can be a useful tool, but not even the best system can beat a skilled human translator. We're just not there yet. If we ever will...

[oss]

in my opinion we will be there in this year. AI is doing crazy things already, which humans can not do. As ai.opensubtitles.com pointed out, for now there is just not so much investment in this area.

[SmallBrother]

in my opinion we will be there in this year.

That's speculation about the future.
Fact is that at this moment AI is not there yet.
So why all the fuzz and all the promotion NOW?
If and when AI beats human beings, THEN you will have my blessings.
But not now...

[oss]

the story is, as always - there are many humans around with different needs. There are a lot of humans, who prefer not-perfect subtitles than no subtitles. So we offer the service and of course we want to be on first rank regarding AI translation for subtitles. so we are getting ready

I get all the points of course, but really, I am into AI, checking what is happening and can tell - we will be there sooner than later. Just recently I make a song using AI

[SmallBrother]

Just recently I make a song using AI

Will you sing that song for us?

[ai.opensubtitles.com]

Because of this, I won't publish the 'upload to OS.org' feature I worked on a month ago

Oh my god. Please don't.

The poetic description of the transcription was more a reaction to these annoying "Why should I pay for anything" comments from the SE salesman, than a realistic review.

Advocate of the devil: His avatar may give the wrong impression, but Jan de Uitvreter is NOT a salesman of SE (even SE being free). Nikse from Denmark is the developer of SE (see www.nikse.dk ). Jan de Uitvreter is just a satisfied Subtitle Edit user. And an excellent subtitler, btw.

I don't think Jan de Utvreter (or anybody) questions the fact that sometimes we have to pay for things. I think, after his findings with the speech to text from SE, his question was legitimate. After all, only ONE error ("Trus" vs. "Truus") IS actually (more than) 99% accuracy.

I'm looking forward to your review of https://ai.opensubtitles.com.

I just posted it: viewtopic.php?f=1&t=1969&start=90#p50183

Since I tested the same models that Jan is so happy about, I can say with confidence that the 99% claim is complete bullshit . I'm testing a lot of languages and AI models and none of them are able or claim to reach 99%.

[hkwill1]

Hi, thanks for the info! AI.opensubtitles.com sounds like a fantastic tool for seamless transcription and translation with the flexibility of choosing different AI models and languages. The credit-based pricing model is a great way to give users control over their usage without worrying about expiration.

[SmallBrother]

Hi, thanks for the info! AI.opensubtitles.com sounds like a fantastic tool for seamless transcription and translation with the flexibility of choosing different AI models and languages. The credit-based pricing model is a great way to give users control over their usage without worrying about expiration.

Hm... Brand new user, 10 months after previous post, only THREE seconds between registering and posting, no added value, just repeating previous info using different words and phrasing, etc. etc...
Looks to me like a bot posting AI generated promotional content.

Dear AI, wouldn't it be better to improve your skills, rather than having AI brothers glorifying your quite mediocre skills?