@nixCraft

I would want an opt in in this case. A normal VLC with the ability to add such a feature-. It shouldn't be forced upon the user.

@nixCraft

mpv is all I need. I still have VLC installed for rare occasions.

@nixCraft VLC said they would be using models on-device, so I don't think there's much to be concerned about other than the extra storage space those models would need? The Whisper model has been an incredible boon for things like podcast transcripts so as long as none of the processing VLC is doing or what you are watching is getting uploaded, this seems like a great feature.

@nixCraft Generated subs are hopefully better than no subs. As long as it’s a pre-trained model that runs locally, and only if explicitly enabled, that’s fine by me.

@nixCraft

I welcome this, I tend to watch all my videos with English subtitles.

English is not my first language so I don't want to miss anything due to not hearing correctly or understanding a certain accent.

At times it is difficult to find proper subtitles, so if the player can provide those automatically, I'm all for it.

@nixCraft
To me the problem is deeper than even the security or privacy. It's that sometimes it's just wrong. Maybe I'm old school but you don't give people wrong data on purpose.

@nixCraft I'm a butlerian jihadist.

@nixCraft I feel like this improves accessibility for people who want to use it. Subtitles can be turned off and then (I hope) the AI wouldn’t even come into play while using the app.

@nixCraft The demo was using offline, local models. So as long as people can turn it off, and any future integration with online services is opt-in, it seems like a reasonable accessibility feature.

@nixCraft I see that current model of AI are not ideal, to say in a lessen way, but in the future it may be used with privacy and respecting data with no so subtle changes.

@nixCraft Did we reach the point where we have to hate everything that is remotely based on the use of neural networks? Am I allowed to use DLSS for image reconstruction, or is that also ethically unacceptable because "AI sucks"? Srsly, isn't there a _slight_ difference between world-destroying plagiarism tools, and borderline magic software like a literal Babelfish?

@nixCraft is it local ai? Or a service? Is it trained on stolen data or paid and open data?

@nixCraft wait what?! I use VLC to avoid all bloat in general. Make it as an add-on if they feel they need to jump on the buzz waggon...

@nixCraft The speech-to-text model runs locally on the user's machine according to this article: https://www.theverge.com/2025/1/9/24339817/vlc-player-automatic-ai-subtitling-translation

@nixCraft This isn't an example of an organization simply giving into the hype, and portraying it as such is dangerously misleading.

This is just speech-to-text: something that has been around for decades, long before the machine learning hype, and is genuinely useful.

Also: Automatically transcribing videos without subtitle metadata is obviously a really cool feature from an accessibility perspective!

Pay attention to what's actually going on before jumping to conclusions and complaining.

@nixCraft I'd be ok with a neural network ai doing work only if it was fully encapsulated on the users' computers via the ai chip or those cpus with ai focus, no outer connections.

@nixCraft As long as you can turn it off, I have no problem with it.

@nixCraft the biggest problem is them calling it "AI". How about locally generated subtitles using a neural network to do the speech-to-text translation?

It is of course a waste of computational resources for 95+% of everything that gets played in vlc, as you can often just download subs. But I can imagine that hearing impaired family members would love this when they watch some home videos of kids birthdays/weddings/graduations/...

@nixCraft
Deaf people are users too. Autogenerated subtitles are better than no subtitle at all.

@nixCraft

That's an accessibility feature. It's needed by some people to watch videos, and VLC's goal it to make everyone able to watch videos. So that feature obviously have its place in VLC.

https://beige.party/@bedast/113799126497343123

AI excels at data classification tasks. There's no denying that. Speech recognition is a data classification task. It's been shown many times that neural network based algorithms outperform the more "traditional" ones at this (and speech synthesis as well). So please stop. It's not generative AI after all.

@nixCraft if the AI runs 100% offline on device, is open source and can be turned off I don't have any problem with that.

@nixCraft Some automated AI does have a practical application, voice-to-text recognition is one of them.

The trouble is these days you never know if AI tools are being ethically applied to improve user experiences, or being shoehorned in so that people can profit off the user's data and make the user's experience worse.

When I see AI is in a product I use, I look for ways to turn it off. Absolutely sick of it being crowbarred into every conceivable thing.

@nixCraft my concern is what data did they use for AI training. The model can be open source, sure, but if doesn’t mean much if the data is from somewhere else, it’s still using other people’s work

@nixCraft Sinceramente, la IA ne gusta y hasta cierto punto es útil. Pero que la metan hasta en la sopa ya me está indignando mucho

#ia #inteligenciaartificial

@nixCraft A VLC plugin for automatic captioning done on device is a great accessibility option. The benefits completely outweighs the issues on GenAI in this case.

@nixCraft AI in any program is horrible. Just another tech "buzz" word to make something seem better and more advanced than it really is.
Auto subtitle generation is bad anyway, just look at youtube.

@nixCraft Is it actually ai or is it another algorithm being called ai because idiots don't know the difference?
Auto captioning algorithms have existed for well over a decade.
What are they *actually* implementing?

@nixCraft I don’t really see an issue with speech-to-text run locally in itself.

An issue however is them labelling it as AI and most likely adding it specifically so they can say they have AI features…

Also, unless they know or really believe the majority of played media is home video rather that movies and TV shows this is the wrong initial approach. There are several online sources of subtitles, better to automatically grab a file from them (like Plex can). Speech-to-text should be the fallback option.

@nixCraft The AI is local, no data is transmitted.
It's a feature I would use.
In fact, I already used a more basic version of it: I used whisper.cpp to automatically generate subtitles for a video that would be played without audio.

@nixCraft
If they do, VLC will likely use live captions, which is an on-device and free open-source program to do human speech to text.
It is in fact so good that they even made a voice keyboard for your phone which runs locally and has generated the rest of this entire toot like everything above and this paragraph without me having to edit basically anything of it.