The short version: a work-from-home agent's call audio is the product of three things — the room they sit in, the settings on their machine, and the software layer cleaning the signal. Teams that sound consistently professional don't get there by luck or by shaming agents in QA reviews; they publish a written audio standard, supply the software layer, and treat household noise as a fact of life to be engineered around. This page gives you the standard, the settings, and a household-noise policy you can copy verbatim.
Vendor pricing referenced below was verified against krisp.ai on . FloorTone covers software and settings only — we don't review or recommend hardware, and this guide assumes whatever headset your program already issues.
Publish a standard: minimum, expected, ideal
The single highest-leverage move is writing down what "acceptable audio" means before anyone is coached on it. Ours has three tiers, and the difference between them is deliberate: the minimum is enforceable, the expected tier is coachable, the ideal tier is aspirational and never punished.
Minimum
Enforceable. A condition of remote eligibility.
Expected
Coachable. A coaching topic, never a write-up on first instance.
Ideal
Aspirational. Recognition, not requirement.
| Tier | Environment | Signal | Enforcement |
|---|---|---|---|
| Minimum | A door that closes during shift hours; no TV/music in the same room while on calls | Caller hears no sustained third-party speech; agent audible without straining | Condition of remote eligibility, stated in the remote-work agreement |
| Expected | Household informed of shift blocks; noise events handled with mute + brief apology, not talk-over | Noise-cancellation layer active on every call; mic at consistent distance | Coaching topic, never a write-up on first instance |
| Ideal | Dedicated workspace away from kitchen/living areas; shift scheduled around known loud windows where possible | Clean QA audio scores three months running | Recognition, not requirement |
Notice what's absent: no purchase requirements. If the program demands specific equipment, the program buys it — an audio standard that quietly taxes agents' wages is a resignation letter generator.
Environment: negotiate with the household, not against it
The environment fix that works is scheduling, not scolding. Most household noise is predictable — school runs, meal prep, deliveries, the neighbour's renovation hours. Ask agents to map their home's loud windows once, then do what WFM does best: put breaks and offline work inside those windows where the schedule allows. It costs nothing and removes the worst collisions. (Our night-shift crowd has the inverted version of this problem — keeping calls quiet at 3 a.m. is its own art.)
For everything unpredictable — the dog, the doorbell, the toddler jailbreak — the answer is a two-second mute reflex plus a software layer that removes the sound the caller would otherwise hear. We wrote up the pet chapter separately in The dog on the call, because it is genuinely the most common WFH noise event and deserves policy language, not eye-rolling.
Settings: the boring 20 percent that does 80 percent
Keep machine settings platform-agnostic and image them if you can. The rules that matter:
- One processor per direction. If a dedicated noise-cancellation layer runs, switch off the softphone's own suppression and any OS-level enhancement. Stacked processors produce the "underwater voice" QA keeps flagging.
- Fix the input device. Lock the softphone to the virtual device (if you run a software layer) or to the headset mic explicitly. "Default device" is how calls end up on laptop mics after every OS update.
- Set gain once, by test call. Two recorded test calls — one quiet room, one deliberately noisy — beat any meter. Keep mic distance consistent; distance drift is the most common cause of week-three quality decay.
- Kill notification audio on the work profile. The caller can hear your pings. All of them.
The software layer: who pays, and for what
If you run a team, the software layer should be a company deployment, not an agent expense. The going rate for the dedicated call-center tier is public: Krisp's CC Core is listed at $15 per agent per month billed annually (verified 2026-09-20), covering agent-side and customer-side noise cancellation plus voice isolation — the full breakdown is in our Krisp for call centers guide, and the strategic picture in the main ops guide.
If you're an individual agent buying with your own card — common among freelance and gig-platform agents — Krisp's individual Meeting plans are the relevant aisle: as of 20 September 2026, a 7-day free trial, then Core at $16/month ($8/month billed annually). One honesty note: Krisp offered a free daily usage allowance for individuals for years, but as of our verification date the pricing page leads with the 7-day trial instead — so check the current terms yourself rather than trusting older articles (including anyone else's from 2025). Before spending anything, also ask your employer: many programs will license a tool that measurably cuts their QA escalations.
The household-noise policy, ready to copy
This is the paragraph we'd put in a remote-work agreement. Steal it, adapt it, and note the tone — it assumes good faith, because policies that assume bad faith create it:
Copy-ready clause for a remote-work agreement
"We know home is a real place with real people in it. We ask three things of every remote agent: a closable door during scheduled call blocks; no third-party media (TV, music, video) in the room while on calls; and active use of the company-provided noise-cancellation software. Occasional household noise is expected and never disciplined — mute, handle it, and come back. A pattern of sustained background noise across multiple shifts is a scheduling conversation first, and an eligibility conversation only if scheduling can't solve it. Your household's noise is our shared engineering problem, not your personal failing."
Two implementation notes. First, the "company-provided" clause is load-bearing: requiring software you don't provide is the same wage-tax problem as requiring equipment. Second, the escalation ladder (schedule → eligibility) keeps team leads out of the impossible position of adjudicating whose baby is too loud.
The 60-second pre-shift check
- Headset plugged, softphone shows the right input device
- Noise-cancellation layer on and processing (most tools show a live indicator)
- Door closed, household pinged on shift start
- Notifications silenced on every device within mic range
- One breath. The queue can wait sixty seconds; the caller can't un-hear a chaotic first impression
Team-lead questions
Can we require agents to buy their own noise-cancellation software?
You can, but you shouldn't. It's a per-seat business expense with a published market price; pushing it onto wages breeds resentment and patchy adoption, and patchy adoption erases the QA benefit you bought it for.
What about agents on very old laptops?
Real-time audio AI costs CPU. Pilot the software on your oldest supported hardware before rollout, and put a minimum spec in the remote-work agreement going forward — machine requirements are standard remote-work practice.
How do we verify the setup without surveilling agents' homes?
Measure the output, not the room: QA audio scores, repeat-request rates, and the software's own usage analytics. What the caller hears is the deliverable — see our audio quality metrics guide. Nobody needs a webcam tour of anyone's flat.