Customer support representatives and remote teams routinely battle ambient disruptions—from bustling call center chatter, barking dogs, and mechanical keyboard clicks on the agent's side, to roaring airport concourses, traffic, and crying children on the customer's end. Left unmitigated, this background noise inflates Average Handle Time (AHT), degrades First Contact Resolution (FCR), and strains customer patience.
Krisp solves this challenge through a bi-directional AI audio processing pipeline: it filters outbound noise from the agent's microphone, purges inbound ambient clutter from the caller's audio stream, and strips both acoustic and room echoes in real time.
Configuring Krisp to achieve crystal-clear audio during live calls requires setting up its virtual audio driver as a smart intermediate bridge, selecting the correct hardware endpoints, and disabling conflicting native audio enhancements in conferencing and contact center software. Here is the comprehensive configuration playbook for frontline agents, distributed teams, and contact center administrators.
How Krisp’s Virtual Audio Architecture Works
To configure Krisp correctly, agents must understand the three-tier routing model Krisp uses to intercept and process audio locally on the workstation:
[ Physical Headset Mic ] ---> [ Krisp App: Local AI Denoising / De-echo ] ---> [ "Krisp Microphone" Virtual Device ] ---> [ Softphone / Meeting App ]
|
(Live Call)
|
[ Physical Headset Speaker ] <-- [ Krisp App: Inbound AI Denoising / AEC ] <-- [ "Krisp Speaker" Virtual Device ] <---------- [ Softphone / Meeting App ]
- Hardware Layer: The agent’s physical USB/Bluetooth headset, analog jack, or external microphone and headphones.
- Krisp Virtual Engine Layer: The desktop application that runs on-device deep neural network models (consuming minimal CPU and introducing only 5–20 ms of latency, with zero audio sent to the cloud for real-time denoising).
- Application Layer: The softphone or contact center software (Genesys Cloud, Five9, NICE CXone, Talkdesk, Zendesk Talk, Salesforce Service Cloud, Dialpad, Avaya) or unified communications tool (Zoom, Microsoft Teams, Google Meet, Slack, Cisco Webex).
Instead of sending physical hardware signals straight to the calling application, audio flows through Krisp's virtual devices: Krisp Microphone for cleaned voice input and Krisp Speaker for cleaned customer playback.
Step-by-Step Configuration Guide for Representatives and Remote Agents
Step 1: Bind Physical Hardware in the Krisp Desktop App
- Launch the Krisp desktop app on Windows or macOS.
- Under the Microphone section:
- Select your physical hardware microphone or headset from the dropdown menu (e.g., Jabra Evolve2, Poly Blackwire, or Logitech Zone).
- Toggle Remove Noise (or Cancel My Noise) to ON. This engages outbound AI noise suppression to scrub background office chatter, fan hum, keyboard clatter, and home acoustics from your speech.
- Under the Speaker section:
- Select your physical hardware headphones or audio output device from the dropdown menu.
- Toggle Remove Noise (or Cancel Others' Noise) to ON. This engages inbound AI filtering to scrub whatever noise exists in the customer’s remote environment before it reaches your ears.
Important: Do not select "Krisp Microphone" or "Krisp Speaker" within the Krisp app itself. Inside the Krisp app, always point to your actual, physical audio hardware.
Step 2: Route Softphone and Calling Software to Krisp Virtual Devices
For Krisp to clean audio in both directions, your calling application must explicitly route its input and output streams through Krisp's virtual drivers.
- Open the audio settings of your softphone or meeting app (e.g., Zoom Settings > Audio, Microsoft Teams Settings > Devices, or your CCaaS browser softphone audio preferences).
- Microphone (Audio Input): Select Krisp Microphone.
- Speaker (Audio Output): Select Krisp Speaker.
Why You Must Select Both "Krisp Microphone" AND "Krisp Speaker"
Many agents mistakenly configure only the Krisp Microphone, thinking Krisp is strictly a mic-filtering tool. Omitting Krisp Speaker causes two major operational breakdowns:
- Inbound Denoising Fails: If your softphone outputs directly to your physical headphones, audio bypasses Krisp entirely, forcing you to listen to the caller's background honking, coffee shop noise, or wind.
- Acoustic Echo Cancellation (AEC) Breaks: Krisp's acoustic echo cancellation algorithm needs to monitor the incoming speaker stream to recognize what sounds are playing out loud. If the software output is not routed into Krisp Speaker, Krisp cannot mathematically cancel speaker bleed leaking back into your microphone.
Step 3: Turn Off Conflicting Native DSP (Prevent Double Processing)
The single most common mistake in remote team audio setup is leaving platform-native noise suppression enabled alongside Krisp. When two AI algorithms attempt to scrub the same audio signal in sequence, they conflict—clipping natural speech frequencies, dropping word endings, and creating robotic, "underwater" vocal distortion.
Before taking live calls, adjust these settings in your calling platforms:
- Zoom: Go to Settings > Audio. Set Suppress background noise to Low or disable aggressive filtering. If available, enable Original sound for musicians or disable high-tier background suppression.
- Microsoft Teams: Go to Settings > Devices. Under Noise suppression, switch the setting from "Auto" or "High" to Low or Off.
- Google Meet: In Settings > Audio, toggle Noise cancellation to Off.
- Contact Center Softphones (Genesys, Five9, NICE, Talkdesk): In the browser or desktop dialer settings, turn off native browser-level AGC (Automatic Gain Control) or platform noise gates where exposed.
By offloading all voice isolation to Krisp's specialized local models, you maintain clean, resonant speech without artifacting.
Configuring Echo Removal: Acoustic Echo vs. Room Echo
Krisp tackles two distinct acoustic problems during live interactions:
1. Acoustic Echo Cancellation (AEC)
- The Issue: Audio coming out of your speakers or high-volume open-back headphones gets picked up by your microphone and sent back to the customer as an annoying, delayed loop of their own voice.
- Configuration: Krisp's AEC is embedded and activated by default whenever both Krisp Microphone and Krisp Speaker are active. No manual slider or toggle is required. As long as Krisp sees the incoming stream in Krisp Speaker and the outgoing stream in Krisp Microphone, it dynamically subtracts speaker audio from the microphone channel.
2. Room Echo Cancellation (Dereverberation)
- The Issue: Remote representatives working in sparse home offices, bedrooms with hardwood floors, or rooms with high ceilings suffer from sound waves bouncing off bare walls. This makes the agent sound distant, boxy, or like they are speaking inside an empty tiled bathroom.
- Configuration: Room Echo Cancellation runs automatically inside the Krisp Microphone outbound stream whenever noise removal is active. It identifies late acoustic reflections and dereverberates the voice signal in real time, transforming a reflective home office into an acoustically treated studio environment without requiring physical foam panels.
Operating System and Browser Nuances
Setting System-Level Defaults
To prevent new browser tabs or softphones from reverting to hardware devices:
- Windows 11 / 10: Navigate to Settings > System > Sound. Set the default Output device to Krisp Speaker and default Input device to Krisp Microphone. In Volume Mixer, ensure specific mission-critical apps (like your CRM dialer) are pinned to Krisp devices.
- macOS: Go to System Settings > Sound. Under Output, choose Krisp Speaker; under Input, choose Krisp Microphone.
Browser-Based Softphones (WebRTC)
Many support platforms (Salesforce Omni-Channel, Zendesk Talk, HubSpot Phone) operate directly within web browsers:
- Google Chrome / Microsoft Edge: Fully supported. When the browser requests microphone access, select Krisp Microphone. In the site permissions for your dialer, confirm both microphone and sound output are routed to Krisp virtual devices.
- Mozilla Firefox: Firefox does not provide granular web application speaker output selection and relies strictly on the OS default output.
- Apple Safari: Due to macOS sandbox restrictions, Safari does not allow web apps to select a virtual speaker device. For contact center operations, teams should standardize on Chrome, Edge, or dedicated desktop client apps.
Enterprise Administration: Centralized Controls and Tuning
For contact center supervisors and IT administrators managing hundreds of distributed agents, Krisp provides centralized policy management through the admin portal (account.krisp.ai):
- Customer Noise Cancellation Sensitivity Tuning:
In Team Settings > Products > Noise Cancellation, administrators can regulate the intensity of inbound caller noise removal:
- Full: Completely eliminates all ambient background sounds from the caller's line. Ideal for high-volume transactional queries where clear intelligibility is paramount.
- Low / Moderate: Filters intrusive noise while preserving subtle auditory context cues (such as whether a customer is roadside, in an emergency environment, or playing a verification tone).
- Mass Deployment via Silent Installers: IT teams can deploy Krisp across Windows (.msi) and macOS (.pkg) fleets using enterprise endpoint managers (Microsoft Intune, Jamf, SCCM). Group policies can lock Krisp virtual device selections so remote agents cannot accidentally deactivate protection during shifts.
- VDI & Virtual Desktop Support: Krisp features dedicated client optimizations for Citrix Workspace, VMware Horizon, and Amazon WorkSpaces, allowing local offloading of audio processing to client thin clients to preserve VDI host CPU.
- Data Privacy & Compliance: Because Krisp performs voice isolation and de-echoing 100% locally on the device, customer and agent voice streams are never transmitted to third-party cloud servers for audio processing. This architecture guarantees full compliance with enterprise PCI-DSS, HIPAA, SOC 2 Type II, and GDPR requirements.
Quick Pre-Shift Audio Verification Checklist
Before taking the first live queue call, agents should execute this 30-second pre-shift check:
| Step | Target Setting | Validation Check |
|---|---|---|
| 1. Krisp App Input | Physical Headset Mic | Speak aloud; confirm the green volume meter bounces. |
| 2. Krisp App Output | Physical Headset Speaker | Click "Test Sound"; confirm audio plays in your ears. |
| 3. Krisp Toggles | "Remove Noise" ON for both | Verify green toggle switches are active on mic & speaker. |
| 4. Calling App Devices | Krisp Mic + Krisp Speaker | Confirm softphone device fields display "Krisp" virtual devices. |
| 5. Native Platform DSP | Built-in suppression Off/Low | Prevent audio clipping and robotic voice degradation. |
Summary and Practical Recommendations
Bi-directional AI noise cancellation fundamentally alters the economics of customer support and remote operations. By filtering audio at both ends of the wire, support agents experience lower cognitive fatigue, callers perceive higher brand professionalism, and average handling times drop because neither party has to ask, "Could you repeat that?"
For individuals and lean remote teams, Krisp's standard plans provide immediate out-of-the-box isolation for popular platforms like Zoom and Teams. For large contact centers and BPO operations, Krisp's enterprise tier—paired with administrative policy locking, inbound sensitivity tuning, and VDI support—delivers a standardized, studio-grade acoustic boundary for every representative, regardless of where they work.
References
- Voice AI for Meetings: Noise Cancellation & AI Note Taker | Krisp
- World’s #1 Noise Cancelling App | Background Noise Removal in Meetings
- Krisp | Call Center Noise Cancellation
- Real-Time Voice AI for Contact Centers | Krisp
- Krisp Pricing Plans
- Integrations Archive | Krisp
- How to Remove Background Noise: Full Guide (2026)