https://semantichearing.cs.washington.edu/

@VivaLaPanda Looks like this can resolve YES as per Rick's provided example.

@IsaacKing hmm, that's not quite to the level I had intended (given that it only does the voice of the person right in front of you, not all voices), but it does seem to probably meet the resolution requirements. I'm leaning towards resolving 90% yes or something to account for the fact the resolution is a bit ambiguous. Thoughts?

@VivaLaPanda It distinguishes the voice in front of you from all voices?

@IsaacKing it filters everything except the voice of the person in front of you (by guessing the location of sound sources)

https://www.macrumors.com/how-to/enable-conversation-boost-airpods-pro/

https://www.wired.com/story/apple-conversation-boost/

@VivaLaPanda That seems pretty selective to me?

If you disagree, this certainly shouldn't resolve to 90% now, since it was supposed to be open for 2 more years, and could resolve fully to YES upon a future product release. You should only consider resolving it to 90% at the proper market close date.

@IsaacKing makes sense to me!

I'll leave this open, but add a thing to the description about the resolution stuff. Thanks for the input ^_^

FWIW, the current functionality of Airpod Pros seems to me like sort-of-but-not-quite what I thought the description was going for, and the conclusion you came to here seems pretty reasonable to me.

The current generation of Airpods Pro do similar things, but I'm not sure how much selection they can do beyond volume, location of source, and frequency: https://support.apple.com/guide/airpods/customize-transparency-mode-dev966f5f818/web

Anecdotally, they seem to do a decent job of, for example, reducing broad-spectrum noise on an airplane while passing voices. (They do this well enough that the first time I used them on an airplane, I was surprised/confused that I could hear the person next to me clearly when I thought I had turned on full ANC.)

Most of my optimism about this comes from:

1) They already sort of do this
2) There probably are categories of sounds that are easy to select out and cancel, even if there remain harder problems like blocking advertisements without blocking conversations.

Most of my pessimism comes from:

1) Tech advances are hard to predict
2) I don't expect Apple to come out with a new headphone chip before 2025

This is going to be much harder. In particular the frequency range of the human voice is very similar to that of many musical instruments (by design), so it takes a lot of additional processing to tell the two apart. Same goes for actual human voice vs human voice heard on a tv.

Does it have to be good at it? Or is it enough if there's a feature that's advertised as doing this?

@IsaacKing doesn't have to be that good, it's enough if it's advertised and semi-functional. I'll use my judgement somewhat, the core concept here is whether things are moving that direction and the technology is improving

I don’t think people are considering compute for partial sound filtering. Since many of these filters might require ML driven Audio segmentation, I’m not sure if they will have nearly enough compute. Maybe in AirPods Max, but very unlikely in vanilla AirPods or AirPod pros

@GabrielAsher I agree. The batteries are tiny, so the available dsp power is also tiny. Anything more then bandpass would require something like an analog NN chip to fit in the power envelope.

@GabrielAsher yes agreed. If this happens by 2025, I expect it'll be on account of one of these:

1. Scalable analog NN silicon

2. Extreme-low latency wireless audio standards so the DSP can happen on the phone (which would be cheating since it probably wouldn't work in settings where a lot of wireless audio products are in use, like airplanes)

3. A resurrection of wireless earbuds with a wire that goes behind the neck, which allows for a lot more battery in the earbuds.

All or nothing active noise cancelation is pretty straightforward, but filtering starts to get complicated. Voices vs ambient noise can probably be done with a band pass, but voices vs advertisements would require lots of slow, power-intensive computation.

"Something in that category" includes voices, so I'll take some YES, but I think only the simplest filters will be viable for the near future.

@citrinitas I don't think band pass would be sufficient for only voices. But yeah, agree generally with the difficulties.

@citrinitas Sony introduced a "focus on voice" band-pass option for ANC on truly wireless earbuds in 2019 with the WF1000XM3. If that satisfies the conditions of this market then it might as well be resolved "yes" right now.