Hector Martin

I’m not trying to record a hit single here. Sometimes people are in the same room as each other.

Physics is physics regardless.

The physics says the microphone converts sound to electricity. It doesn't know nor care how many people are speaking at once or what their ranges are, as long as they're within the mic's response. Cut your bullshit. The algorithms suck. Physics has nothing to do with it.

Yep, but the algorithms try to reduce noise. If one person's voice has more overlap with the noise than the other, it'll adversely affect their speech. The reason for such algorithms...

...is because our ears are really good at using spatial release from masking. We localise signals and noise and hone on the areas where the signals come from. We can't do this when listening to a mono feed from an omnidirectional microphone. Because physics & psychoacoustics.

That's all fine and well, but if the algorithms bias for men and against women, that's a bug that needs fixing and suggests biased testing. We're not talking about corner cases where there is a lot of higher frequency noise that would overlap more with female voices here.

Not necessarily. I can't speak for the engineering side, but in psychoacoustics there's a big effort to account for sex in experiments to prevent bias...

...It's might not be that the algorithm is biased against women, but that higher voices (women and children) are more similar to the noises in homes/offices that are being filtered out. Sure it sucks (and could be improved), but I wouldn't assume a testing bias is the cause.