An interesting angle to this: if your phone is powerful enough to transcribe audio in real time without any external processing, and considering how cheap it is to transfer and store text, how long before autocratic countries start persisting everything you say ?

Before or how long time ago, every countries do that?

I wish mere mortals could have access to this. Having an easily searchable logs of what was said on calls would solve trillions in contract and court disputes. Many people already use this with third party apps, especially on Android.

Persisting is benign. Actively processing, less so. Imagine stack-ranking party alignment sentiment, based on everything the subjects ever said.

The processing part was implied.

My point was just: with audio, you might want to think about really recording, storing and processing everything. But if a phone can do high quality transcriptions and just send them off to government data centers, it becomes sort of a no brainer.

Also available on APKMirror

https://www.apkmirror.com/apk/google-inc/google-recorder/

Loving the trend towards federated on-device ML. The more data that stays on-device rather than going to Google/whoever's cloud the better.

I would't put it past the cloud providers to ALSO snarf up the transcript (and possibly the raw audio to help with training).

It might even be a cost savings.

If they're federating the ML, they can also do federated learning. (https://en.wikipedia.org/wiki/Federated_learning) Although in most cases I suspect it's probably not federated learning, instead just a deployed model that they've trained with in house data.

so, where can I download the app for a different phone? I'd like to record to text my day on a phone that has no internet. I can't see it on the "play" store

I think it needs a special AI accelerator chip which is first coming in the Pixel 4 phones.

I don't care if it takes a few minutes to process the data... I just want the app

so the chances of it on the iPhone are slim then :(

Apple would more likely to remove every single feature from iPhone before allowing something like that on phone calls.

You mean before allowing someone else to do it? You almost sound like you think apple cares about privacy any more than it serves their financial interests at the moment.

I thought apple already had ML functions like this for facial unlock.

They also used to provide transcripts of voicemail done locally on the phone (although it was not great accuracy)

They also use ML for "Hey Siri" and the face detection in the photos app.

Edit: Here is the article where apple talks about how "Hey Siri" works: https://machinelearning.apple.com/2017/10/01/hey-siri.html

Google: totally not an APT

on a somewhat related note, I've discovered that youtube is somehow exfiltrating conversations I have during the day, to tailor content on my YT feed..

There does not seem to be any one app responsible for the data exfil, so I've been led to believe they're building models (or transcribing text locally) and sending it up whenever android phones home (constantly).

I've tried attempting to block data exfiltration (Netguard etc) but nothing seemed to have worked and I had to switch devices.

If they're able to do on-device transcriptions and the tech is mature enough that they're going to release it as a standalone product with a big launch, I can now believe they have some similar tech running on devices to support their core-metrics (engagement w/ YT content, which drives Ad revenue).

Play close attention to your YT video feeds and you may notice the behavior I've observed too.

I'm not going to say it's not technically possible, because it is (although it would be a huge battery or bandwidth drain, not to mention an insane tech scandal if found to be true); but if the only proof you have is a video feed that resembles your conversations I'm going to guess you're giving that data to Google is some other way. There are a thousand ways this could be happening, and most them have a psychological component.

I've made significant efforts to minimize dependency on google services and these are items that appear w/o any interactions or data being explictly shared to the google platform. It was hard to believe at first, but after quite a bit of testing I realized what was happening.

Have you considered that this traffic is impossible to detect or block because... it doesn't exist?

there's traffic that gets through services like netguard that's calling home to services like mtalk.google.com and google play. This traffic comes up when doing router level DNS & traffic monitoring.

this has also been observed on an AOSP device that does not have Gapps installed.

They aren't. You are noticing videos that have a similarity to things you have talked about recently. It's the same phenomenon that occurs when you learn a new word, then all the sudden it starts popping up everywhere.

The new word thing always amazes me.. you swear you never heard it before and invariably it shows up once or twice in the next few weeks at a minimum.

This is also a great talk on the magic of science and how magicians exploit flaws in our brain or sensors (e.g. the eyes) perceive and process the environment. One of his examples is that when your eyes are moving from one focus point to the next you are literally blind and don't see anything (basically to prevent motion blur) - but your brain fills in the missing info, and so you can hide sleight of hand in these moments. One of many examples: https://www.youtube.com/watch?v=EwuRSqODRdA

There are also other explanations for this advertising thing such as you thinking about it because you saw it somewhere, having actually searched or clicked on things related to it and even if you talk to someone else and THEY search about it from the same IP address it can also end up in advertising feeds. And i'm sure many more.

There’s a name for it- Frequency illusion or Baader–Meinhof effect: (https://en.m.wikipedia.org/wiki/Baader%E2%80%93Meinhof_effec...)

No. This is not the case. It's not a new word or idea, it's a topic and content blocks that appear. Including items discussed in conversation appearing as a recommended video, despite being uploaded years ago. These are ephemeral recommendations cards that dissappear if there's no engagement at the next app launch.