Live data from Hacker News

AI headphones let wearer listen to a single person in a crowd by looking at them

washington.edu

61–70 of 452 posts

Re: AI headphones let wearer listen to a single person in a crowd by looking at them

#62

This but more advanced would quite nicely help with my tinnitus. I hear fine when one person is speaking (even softly and at a distance), but multiple or with music, I hear nothing.

In the same boat. I have some tinnitus (low frequency, radio static like noise) and struggle with conversation in loud places with lots of background noise and conversation. If I sit in a loud bar it is hopeless hearing what anyone but the closest two persons are saying. Conversation in normal settings are mostly no issue.

So something that would enhance the speech of whomever I'm looking at would be super cool. Apple AirPods already have some sound shaping abilities to react to environment and mode. They also support specific voice enhancement if you put your phone down in front of the person speaking. If they ever support directional voice enhancement, like in this research, directly in the AirPods it would help me so much with social interactions in loud places.

Re: AI headphones let wearer listen to a single person in a crowd by looking at them

#63

How much is the AI necessary for this? At least for the targeting of sounds in the line of sight, that should be fairly easy to do without AI, but I don’t know about the human voice identification.

I think one could build quite the good system with 2 directional microphones and then do some beamforming or how it is called to isolate the depth one want to perceive.

But this is super expensive since you need calibrated mics etc.

The biggest advantage of neural nets in this field is that you can use a dirt cheap microphone and postprocess it so good that it is good enough or even very good for humans.

Re: AI headphones let wearer listen to a single person in a crowd by looking at them

#66
This remembers me of NVIDIA RTX Voice [0]. Although not made to isolate single persons, this is quite impressive. I hope that this single person isolation will find it's way to consumer noise-cancelling headphones

[0] https://www.youtube.com/watch?v=uWUHkCgslNE

Re: AI headphones let wearer listen to a single person in a crowd by looking at them

#67
post #56

They couldn't use this to listen to me. They would just get "I am just a large language model, I can't help you with that." I use a lot of curse words. ;)

I asked gpt to translate for me the lyrics of a recent popular song containing the word "puta" and it just refused "I'm sorry, I can't help you with that". When I insisted it just ended the conversation.

YouTube's automatic captions do this for any word on an ambiguous blacklist. Not only is this annoying when reading the captions, but I imagine that for the hearing impaired, it would also be condescending to have Google tell you what you can and cannot read, especially since the hearing un-impaired experienced it just fine.

Re: AI headphones let wearer listen to a single person in a crowd by looking at them

#68
post #38

Earlier quoted context omitted.

And we almost all will be where you are now, if we live long enough. If you can pick out audio from individuals, you could also send it through speech recognition and subtitle real world conversations for when hearing is worse or not there at all. But it really needs mobile devices capable of doing the processing locally, as I think round trips to the cloud would make it less useful or potentially useless.

How much latency is acceptable? If you're off in the woods somewhere far from the cloud, sure, but it's less than 10ms to ping Google.com for me, and if the speech-to-text engine runs faster than realtime, I don't see why processing remotely is a problem. 10 ms is nothing. Still, the transcription part is already here today. The Google Translate app has a transcribe app that does this (runs locally; does not do magic…

> If you're off in the woods somewhere far from the cloud, sure, but it's less than 10ms to ping Google.com for me

I'm in one of the biggest cities of my current country, and the RTT to google from me is 87-91ms. Well over 4 million people live within 100km of me, so I suspect they see similar latencies. On my cell, I see 191-207ms.

Re: AI headphones let wearer listen to a single person in a crowd by looking at them

#69
post #67

Earlier quoted context omitted.

I asked gpt to translate for me the lyrics of a recent popular song containing the word "puta" and it just refused "I'm sorry, I can't help you with that". When I insisted it just ended the conversation.

YouTube's automatic captions do this for any word on an ambiguous blacklist. Not only is this annoying when reading the captions, but I imagine that for the hearing impaired, it would also be condescending to have Google tell you what you can and cannot read, especially since the hearing un-impaired experienced it just fine.

And then there's TikTok who will downrank your content if its AI catches you using words like "kill" or "suicide" [1]... and so, as many creators cross-publish on YouTube Shorts as well, it also automatically degrades the content there as the creators use "euphemisms" instead. And then young people snag that up and literally write that way on Reddit.

No thanks, I don't want to hear or read "unalived" again. And for fucks sake it's high time the US government steps in on Tiktok - when the CCP literally influences how our children speak it's gone way too far!

[1] https://www.wired.com/story/algorithms-suicide-unalive/

Post reply on HN