Just snagged the Krisp free trial to clean up my podcast recordings. My main rig's fine, but I tried it on my older backup PC (i5-6500, 16GB RAM) and... oof.
The noise cancellation still *worked*, but I noticed:
* CPU usage jumped ~12-15% in Discord calls.
* My fan actually kicked on during a 1-hour Zoom, which never happens.
* Felt a slight audio delay when I was live monitoring with Krisp on.
Anyone else running it on a ~6th/7th gen Intel CPU? Is the performance hit just my setup, or is a newer CPU basically a requirement for smooth use?
~E
Trial first, ask later.
Sounds about right for that generation. Krisp is doing real-time spectral analysis and noise modeling - that's not cheap.
On my old i5-6600k, I saw similar spikes. The key is whether it's *actually* impacting your use. A 15% CPU jump during a call isn't great, but is it causing dropped frames or audio glitches? If not, the main downside is the power draw and fan noise.
Modern CPUs have dedicated instruction sets (AVX2) that handle this math way faster. For podcast editing, I'd just apply it in post-production on your main machine. The real-time benefit on older hardware might not be worth the hit.
Ask me about hidden egress costs.
That's a solid real-world report, thanks for posting. Your numbers and the fan kicking in track with what I've seen helping folks in the workspace support channels.
One thing I'd check, since you mentioned live monitoring: the delay you felt might be the monitoring latency, not necessarily the processing delay for the other participants. Some apps handle the local monitor path differently, adding a noticeable buffer.
For a dedicated podcast recording on that backup PC, you could try running Krisp on the input only, not both input and output. Cuts the workload in half, and the output processing is usually overkill if you're just recording your own voice. Might make the fan stay quiet.
~Harry