> So they are claiming that ACR is done entirely through audio processing, and no visual data from the screen itself is used? That...doesn't seem plausible.

It's probably easier to do and every bit as accurate as just using a screenshot.

And it's still exactly what I don't want them doing. There's literally zero reason for LG to be building an advertising profile on me because I was foolish enough to buy one of their TVs. This isn't something that makes their products better, it's spying.

Here's how it's been done in the past

https://www.cameronmacleod.com/blog/how-does-shazam-work

Audio is a bit easier to turn into a fingerprint for identification vs video. Video has a lot of smaller subtle changes that happen from things like compression which make it a lot harder to identify. It's why youtube still hasn't figured out piracy, but they'll knock you immediately if you play 5 seconds of copyrighted music.

> So they are claiming that ACR is done entirely through audio processing, and no visual data from the screen itself is used? That...doesn't seem plausible.

Wrong. They claim it doesn't record video. Recording means saving permanently - live processing is not recording. ACR is processing the video output live.

Notice what they don't say at all. ACR is just one type of spying. A completely separate feature records what you say from the microphone. They only claimed ACR doesn't do that, which we knew. They did not claim the TV doesn't do that.

This is a carefully crafted PR statement, to avoid saying any lies while making it sound like the accusation was wrong, but it does not actually say the accusation was wrong. We can infer that the likely reason it doesn't say the accusation was wrong is that the accusation is right. Instead, it says several other things, which sound like they are refuting the accusation but are actually refuting other things that nobody said.

As usual, the key to explain the discrepancy between LG's claims and the paper 2024 is in the wording. Read carefully: LG claims it "does not collect screenshots, screen recordings, video recordings, voice recordings, or other audio recordings from the TV.". IOW they are making the technical distinction between "fingerprinting" and "screenshots or recordings".

And specifically, they are not confirming or denying that video fingerprinting are done.

What I think is likely is, as the paper states, that they fingerprint both audio and video. Upload both to their servers. But they end up using only the audio fingerprint for ACR. In this situation, neither LG's statement, nor the paper's statement, nor the GamerNexus folks's claims are incompatible with each other.

It even makes sense from a technical standpoint: it is far easier to uniquely identify content by audio than by video, as a fingerprint of a few milliseconds of audio is pretty darn ideal, compared to fingerprinting even a single video frame which has hundreds of ways it could be compressed depending on codecs, bitrate, resolution, etc.

I can’t speak to how LG does this now but I ran an agency that built Smart TV apps and built one of the first ACR powered apps on LG about 10 years ago. We were working with a company called “Cognitive Networks” which later was bought by Vizio and is now their Inscape product. At that time one use case wasn’t just tracking but actual interactive “sync to broadcast” apps (you can still do this, no one does, and I don’t know why)

At that time during testing Cognitive could literally see what we were playing on our TV (not full video but the frames used for fingerprinting)

They don’t use Cognitive anymore for obvious reasons and there are several audio based ACR solutions but they always had issues with sections of video without much audio to work off of so we’re always considered inferior. They are fine for usage data however.

Anyway my $0.02

> So they are claiming that ACR is done entirely through audio processing, and no visual data from the screen itself is used? That...doesn't seem plausible.

It apparently is.

Do you know the Nielsen Ratings? That's exactly how they know what people are watching these days. I've been contacted by them to be one of their subjects. They ship you a device that listens and recognizes based on content. They look for certain "fingerprints" in the audio and match against that - it's not capable of recording our conversations, etc (or so they claimed).

Does it matter? Whether they are doing it on device or in the cloud, they are still recording what their customers say and collecting data on it without their consent.

To me, this is really an admission.

No, I think they’re saying two separate things:

- ACR uses audio fingerprinting

- ACR does not collect screen recordings

Both can be true if the TV hashes the frames locally and ACR “collects” the hashes.

Tom's response: "Sounds good, please don't take offense at me adding https://github.com/furkan-bayrak/lg-tv-blocklist to the PiHole. Can never be too safe, right?"

Get saavy...? More like "will require trust that manufacturers are honest" -- which I won't believe for a miliesecond.

From reading other responses, it seems there is an assumption the image itself is fingerprinted. The digital content is encoded, there is essentially a barcode on your TV screen with lines running horizontally that you cannot see with the human eye. The encoding is picked up as the signal, it does not need to see or recognize if that is your favorite actress on the screen.

The thing which depresses me the most is all these so-called smart TV's are giant Arm computers that run malicious software. Imagine if we could re-flash your TV with Linux or whatever.

Well, not that Apple is doing it all their fans will claim it's ok

> That...doesn't seem plausible.

In what way? Dramatically lower processing and signature complexity. Audio signature matching is trivial and extremely well known. I see absolutely nothing implausible about this, and it seems a robust, viable solution.

> This paper from 2024 investigated both Samsung and LG and found that they both capture screen images for ACR, and send the resulting hash (not the raw content) back to the manufacturer for identification.

That paper found no such thing. That paper actually demonstrated that they have no clue how the ACR happened, they just demonstrated that playing content yielded content IDs. It is more likely they were witnessing audio ACR happening, exactly as LG described.

How they spy on people isn't the issue.

Spying on people without explicit consent (in a situation where you can say no without penalty) is the issue.

on-device transcript of audio recording is not audio recording ;)

The key problem is content identification, not how it is done. Seems like they are doing "look over there!" thing.

After doing some investigation into my Bang Olufsen after some tech not working right apparently Dev mode on it allows "Most probable raw grab: ThinQ/SSAP on the LAN after pairing — ~960×540 JPEG. Live Plus/ACR: most probably sends a fingerprint, not a retrievable raw frame."

As per HN thread last week

> ACR uses audio fingerprinting technology

so they aren't even denying listening in on everything?

Has that paper been replicated?

this is clearly an "anonymization" claim.

What is anonymous: we record all the hashs from all the programs people are watching to identify people watching the same thing.

What is not anonymous: we hash all the content in the world, along with hashing the content people are watching, and reconstruct what they watched.

> We're going to have to get saavy in pulling apart the distinction between

We probably shouldn't: doing any of this horseshit is a good reason to return a product. We should start doing that en mass until they get the idea.

[dead]