[formerly_proven]

"It depends".

Digital built-in microphones use DMIC, which aiui is a one-wire interface where the microphone just sends a delta-sigma bitstream. If you implement the switch through a multiplexer or logic gate this kills the signal 100 % essentially.

Analog electrets can't be just shorted, because that causes a loud BANG when you switch due to the bias voltage, so you use a capacitor in series, which only shorts the AC portion. Because of the impedances involved, this only gives you 40-60 dB of attenuation, which isn't enough for a good ADC.

Similar for XLR microphones (hot+cold are shorted, not disconnected, because of phantom power).

[jrockway]

Why do you need to short it? Just move the wires from "on" to "high impedance" (similar to what a pair of scissors would do when applied to the cable).

[formerly_proven]

The assumption is you want a click/pop-free mute. If you don't, then just SPDT the signal input to ground, problem solved. But if you do want it to be pop-free, you can't be disturbing the DC bias path (as explained above), so that isn't an option.

[amelius]

Can't you switch the bias voltage off?

[anticensor]

It would make the microphone half-off, still capturing the upper half of the soundwaves.

[formerly_proven]

No, the output of electret capsules is generally wired as a common source amplifier at Ugs ~0 V with an N-JFET. Without bias, there will be neglibible output (~essentially only the capacitive coupling from the gate to the output; if you SPDT the bias voltage to ground, you're having a >100 MOhm source impedance (the capsule) fighting a couple kOhms (bias + input resistance) through perhaps 5 pF or so.

[formerly_proven]

Yes, but then we're back to loud popping sounds when you turn it back on.

[amelius]

You could ramp up/down the voltage gradually.

Anyway the thing is connected to an ADC. The computer can sense when the switch is turned on/off, and turn off (or flush) the audio pipeline at the appropriate times.

[formerly_proven]

How does the ADC distinguish the transient from switching the bias voltage from an intentional signal? (Yes, if you see the entire waveform, this is quite easy, but because of the low frequency, this would incur another ~20-50 ms in latency, which is unacceptable).

Ramping the bias voltage requires additional components (cheapest way these days would probably be a separate DAC integrated into the audio codec, but then you are back to not having a physical kill switch) and also incurs extra delay for turning off and on (probably 100-200 ms).