I've recently had to participate in some Zoom calls where some participants were using computer audio and other people were using the phone bridge.

This really highlighted how much more difficult it is for me to understand people over the phone.

(thread)

It took *much* more effort for me to understand what the phone callers were saying than the callers with webcams.

And it wasn't just a side effect of lip reading — even when I looked away from the screen, I felt much less fatigued listening to the webcam callers.

The reason for that? Sampling rates!

In most cases, on phone calls, audio is sampled at 8 KHz.

(In some cases, you *may* get ≥16K if it's a mobile-to-mobile call, both phones and networks support HD Voice, and the providers agree on how to send it.)

Meanwhile, most online voice chat services support audio with a sampling rate of at least 22 KHz - depending on the service, it may even go as far as 48 KHz.

That's significantly higher than the typical phone call. And the difference isn't trivial.

(I was going to post some videos with spectrograms and audio, but Twitter just ate the clips, saying the format was invalid, without recovering the rest of what I'd typed to accompany them. So I'll try this with screenshots instead...)

Update: I figured out a way to encode the A/V clips so that Twitter won't choke on them, so I'll be able to post the rest of this thread the way I originally intended after all!

Here's an audio clip of me saying "sail" and "fail", recorded at 44 KHz, and an accompanying spectrogram with the cursor moving across it as the audio plays.

Note how the /s/ and /f/ are very distinguishable in the spectrogram—but almost entirely above 4 KHz.

In audio encoding, the sampling rate is 2 times the highest frequency that an audio file can reproduce.

So if we record the same audio at 8 KHz, that means those frequencies above 4 KHz disappear.

Note how /f/ and /s/ are *much* harder to tell apart now!