Software vs Room Treatment: The Workshop Verdict
The verdict in one paragraph: the room and the software are different tools for different noise, and the order is settled: physics first — distance, seals, surfaces — because it's permanent and free; software after, for the residue physics can't touch: unpredictable household life, and noise arriving from your callers. As that software layer, Krisp earns 4.4/5 here: virtual mic and speaker under every app, both directions, voice isolation for the voices problem, free plan of 60 min/day (verified July 28, 2026), unlimited at $8/mo annual. It's the right second layer — and a mediocre substitute for a first one.
Add the software layer freeHow the workshop assessed this
From verified vendor facts (Krisp's help-center plan terms and pricing, checked at the date above), the acoustics fundamentals this whole site teaches, and the structural logic of device-layer filtering. No staged decibel charts — the honest test is the A/B at the end, run in your own treated-versus-untreated room on your own household.
The division of labour
| Noise class | Room's share | Software's share |
|---|---|---|
| Echo / hollow voice | All of it — distance + surfaces | Marginal; reverb defeats live filters |
| Steady hum, fans, traffic bed | Most — seals, distance, gain | Finishes the residue cleanly |
| Keyboard & desk transients | Decoupling helps | Strong — trained on exactly this |
| Household voices, pets, sirens | Doors help, furniture doesn't | Its home game — isolation + models |
| Noisy callers (incoming) | Nothing | Exclusive — both-direction filtering |
Scenario read-through, limitations attached
The treated room with a toddler
Rug, curtains, close mic — and a two-year-old. This is software's case: voices are the class furniture can't filter, and enrolment-based isolation plus trained models raise the floor meaningfully. Limitation: a screaming child at one metre still partially lands; the door protocol remains on duty.
The bare rental with a fan hum
Software alone here disappoints — hum plus echo plus distance is the worst input a filter can receive. Limitation (of buying software first): run the free physical layer first; the same subscription sounds twice as good afterwards.
The consultant with loud clients
Incoming cleanup — their construction site, filtered before your ears — is the capability no room purchase provides. Limitation: it cleans noise, not their choppy network; garble isn't in scope.
The old desktop in the den
Device-layer ML bills the CPU. Limitation: aging machines feel it; trial the free plan on your actual hardware before subscribing.
Pricing (verified July 28, 2026)
| Plan | Price | Terms that matter |
|---|---|---|
| Free | $0 | 60 min/day filtering (1.5× with work email), midnight-GMT reset, one device; unlimited transcription, 2 AI notes/day. Perpetual plan, not a trial. |
| Core | $8/mo annual · $16 monthly | Unlimited; the separate 7-day full trial (no card) lapses back to Free. |
Who should skip the software layer
- Echo sufferers — your fix is a hand-span and a rug, not a subscription.
- Quiet-room dwellers — if the loop test already sounds clean, there's no residue to buy a filter for.
- RTX-on-Windows owners — NVIDIA's free Broadcast covers the core case; trial it first.
- Anyone who hasn't run the physical layer — come back after; the same money buys more then.
The A/B that settles it
- Treat the room per the guides; run the voice-memo loop and save it.
- Add Krisp's free plan; record the same sentence with your household being itself.
- Have a colleague — not you — pick the winner on a real call, worst hour of your week.
- Decide with the meter honest: if the room alone won, this page told you to keep your money.
Layer two, priced at zero to try
Past a point, another foam panel buys nothing — the remaining noise is unpredictable life, which is what ML filters eat. Krisp, 7 days full-featured, no card: find out if your room's last 10% is a software problem.
Test the last 10%