You sound great when you are facing your microphone. You turn to your second monitor to read something, keep talking, and the person on the call says "sorry, you cut out."
You did not cut out. You moved about eighteen inches. Three separate bits of physics are conspiring against you, and they compound.
1. Microphones are directional on purpose
Most desk microphones — the Blue Yeti in its default mode, the Shure MV7, essentially every podcasting mic — are cardioid. They are deliberately most sensitive to sound arriving from directly in front, less sensitive from the sides, and close to deaf directly behind.
That is a feature, not a defect. It is why a cardioid mic ignores your keyboard, your fan, and your neighbour's dog. The trade-off is that off-axis sound is genuinely, measurably quieter.
Roughly, for a typical cardioid: about 3 dB down at 90° off-axis, and 15–25 dB down toward the rear. Turning your head 45° while talking is not a rounding error — it is a real level drop before you have moved an inch.
2. Distance punishes you faster than you think
Sound follows the inverse-square law: double the distance, and the sound pressure reaching the mic drops to a quarter — about 6 dB.
People consistently underestimate this because it is not linear. Sitting back in your chair can easily take you from 8 inches to 24 inches. That is a tripling of distance, which is a drop of roughly 10 dB. Your voice is now less than a third of its previous level at the capsule.
And here is the compounding part: turning also increases distance. Rotating toward a side monitor moves your mouth further from the mic and pushes you off-axis. The two effects stack.
3. The room gets proportionally louder
This is the one that actually ruins intelligibility, and it is the least intuitive.
Your direct voice drops off with distance. Reflected sound — your voice bouncing off walls, desk, and monitors — is far more evenly distributed around the room and drops off much more slowly.
So as you move away, the ratio between direct sound and reflected sound shifts. Close up, the mic hears mostly you. Further away, it hears a much larger proportion of room. That is precisely the "talking from the bottom of a well" sound, and it is why simply turning the gain up does not fix it — you are amplifying the room along with the voice.
Speech intelligibility depends heavily on that direct-to-reverberant ratio. Once room reflections start competing with your direct voice, consonants blur, and both humans and speech-recognition models start guessing.
Why dictation degrades before humans complain
People are remarkably good at pulling speech out of noise. Speech recognition is less forgiving. Long before a colleague says anything, dictation will start dropping word endings and mishearing short words — because the acoustic detail it relies on has been buried in room reflection.
If you dictate and find yourself repeating sentences, it is usually not the software being bad. It is that you drifted off-axis and the input quality quietly fell off a cliff.
What actually helps
Get closer. The single highest-leverage change. A boom arm that puts the mic 6–10 inches from your mouth beats almost any processing.
Aim it at your mouth, not your face. Slightly off to one side, angled up or down toward your mouth. This also reduces plosives.
Wear the microphone. A lav or headset mic keeps a fixed distance no matter how you move. Acoustically it is the most reliable answer to this whole problem.
Treat the reflections. Anything soft behind and beside you. A bookshelf or curtain does real work; you do not need studio foam.
Don't just raise the gain. It amplifies room and noise equally. It changes the volume, not the ratio that determines whether you are understood.
Or: stop fighting the geometry
Every fix above asks you to change your behaviour — sit still, stay on-axis, keep a fixed distance. That is fine for a recording session. It is unrealistic for a working day where you are turning between screens constantly.
The alternative is to put a microphone in each place you actually are, and let software pick whichever one currently hears you best.
That is what SoundMesh does. It measures speech against each microphone's own noise floor across up to 8 inputs, and crossfades to whichever is cleanest as you move — so turning toward your second monitor hands off to the mic that is now closest, instead of costing you 10 dB.
Whichever path you take, the underlying lesson is the same: your microphone is not failing. It is doing exactly what a directional transducer does. The fix is either to stay in its sweet spot — or to make sure there is always a sweet spot wherever you happen to be.