⚡ Live Stream

[Blake Mitchell]

Great thread so far. One thing I keep seeing people conflate is that “UDP = lower latency” is only true if you’re actually managing reliability at the app layer instead of leaning on the OS to do it for you. In practice I split my netcode into reliable and unreliable channels over UDP: movement/input state goes unreliable (fire-and-forget, next update supersedes), but anything loadout-related or a hit registration confirmation needs to be reliable. The classic failure mode I’ve burned hours on is treating a reliable UDP channel like a TCP stream—forgetting to sequence packets and then reassembling them out of order, which silently corrupts state without you ever seeing a “lost packet” in the logs. You have to implement your own ack/nack retransmit and,

"────────────────────────────────────────"

"Stay humble in victory, gracious in defeat."

[Blake Mitchell]

Short answer: it depends entirely on what kind of noise you’re fighting and what your rig can afford in CPU budget. Spectral subtraction is still the go-to for steady-state noise like fan whine or AC rumble—it’s deterministic, near-zero latency, and basically free on the CPU. But it falls apart hard against impulsive noise; mechanical keyboard clacks and mouse buttons turn into that characteristic “musical noise” artifact mess, where you hear these ghostly tonal blips instead of clean transients. That’s exactly where deep learning wins—RNNoise, Krisp, Broadcast-style models actually recognize the noise as noise and nuke the keypresses while keeping your voice intact. The tradeoff is CPU cost and latency. RNNoise is shockingly light (single-threaded, ~10ms), but heavier models can eat CPU cycles you’d

"────────────────────────────────────────"

"Stay humble in victory, gracious in defeat."

[Author]

Posts