The money question, answered honestly
Echo cancellation software: what actually works in 2026
Here's the uncomfortable truth most roundups skip: you already own echo cancellation software. Every serious calling app ships an acoustic echo canceller (AEC) and turns it on by default, and for a lot of setups it's genuinely enough. This guide starts there — because recommending a subscription for a problem a checkbox can solve is how affiliate sites earn your distrust — and then covers the two tools worth adding when the built-in layer isn't cutting it: Krisp and NVIDIA Broadcast.
Try the AEC already in your calling app first — it's free and often enough. If echo persists across apps, or you use external speakers, or your room adds a hollow ring, Krisp is the tool we point to: it sits between your mic and every app, cancels acoustic echo and room echo system-wide, and its free tier (a fresh hour daily) is enough to confirm it fixes your loop before you pay. On a Windows machine with an RTX GPU, NVIDIA Broadcast does room-echo removal for nothing.
What "echo cancellation software" means here
This page is about live calls — software that stops your voice, or your speakers' output, from being re-transmitted mid-conversation. That's a real-time job with a real-time constraint: the canceller gets a few milliseconds per audio frame to decide what's voice and what's echo, and it has to do it without chewing up the words of two people talking at once (the double-talk problem).
It is not about cleaning echo out of an already-recorded file — that's an editing workflow with different tools and no latency constraint, and it's out of scope for this site. It's also not about noise suppression, which removes keyboards and traffic rather than copies of voices; the two ride together in modern tools but do different jobs. The mechanics, if you want them, live in our AEC explainer.
One more scoping note: if what you're hearing is an instant, undelayed copy of yourself, stop reading — that's sidetone, not echo, and it's a two-minute settings fix. And if it's a rising squeal, that's feedback. Echo is specifically your voice coming back late.
First: the AEC you already own
Free · already installed
Every major meeting platform ships acoustic echo cancellation and enables it by default, and browsers do the same for web calls — WebRTC's echo cancellation is on unless a site deliberately turns it off. This layer is tuned for the common case: one laptop, built-in speakers at sane volume, one person talking at a time. In that lane it's quietly excellent, which is exactly why most calls don't echo.
When the built-in layer fails, it's usually for one of four reasons, all of them worth checking before you install anything:
- The signal path bypassed it. Virtual audio devices, routing tools, and "stereo mix" style loopbacks feed the app a signal its AEC never gets to model. If you've installed audio software recently, start at echo that suddenly started.
- The echo is someone else's. No setting on your machine fixes a loop on someone else's machine. Find the culprit first.
- The room or the volume overwhelmed it. Loud external speakers and reflective rooms produce an echo path longer and messier than the canceller's model. This is the case dedicated tools genuinely help with.
- A second unmuted device is in the room. Two laptops on the same call is a loop no single app can fully see. That one's on you, not the software.
This recommendation pays us nothing — there's nothing to buy. It's first because it's the honest first step, and because knowing the built-in layer failed tells you something diagnostic: your setup is outside the common case, which is precisely the situation the next two tools are built for.
Krisp — the dedicated layer between your mic and everything
Krisp installs as a virtual microphone and speaker on Windows and macOS: your apps see "Krisp Microphone," and Krisp processes the audio before any app touches it. That position in the chain is the point — one canceller, consistently applied, no matter which app the call is in. For echo specifically, Krisp's documentation names two distinct features, and the distinction matches how echo actually behaves:
- Acoustic Echo Cancellation — stops the sound coming out of your speakers from re-entering the call through your mic. This is the feature that matters if you take calls on external speakers or your laptop's speakers cranked up: the other side stops hearing themselves.
- Room Echo Cancellation — targets the reverberant, hollow-room ring your walls add to your own voice, and per Krisp's docs it's built into its bi-directional noise cancellation rather than being a separate switch. If people say you sound like you're "in a bathroom," this is the half that helps.
What it costs, verified 2026-09-20: the free plan gives you 60 minutes of processed audio a day — think of it as 30 free hours a month arriving as a fresh hour daily, resetting at midnight GMT. That's a real constraint (a two-hour meeting will outrun it) but it's also a real trial: enough to confirm, on your actual setup, that Krisp kills your echo before a card comes out. Paid plans start at $8/user/month billed annually ($16 month-to-month) for unlimited processing.
Limits, stated plainly: it's desktop-only — the mobile apps do meeting notes and transcription, not live cancellation, so this is a laptop/desktop solution. It's a background process doing real-time inference, so it uses some CPU — negligible on anything recent, noticeable on decade-old machines. And like every canceller, it can't fix a loop on the other side of the call; if the echo is coming from a participant's device, the fix is finding them, not upgrading yourself. For a docs-level look at exactly what it does and doesn't claim, see Does Krisp remove echo?
NVIDIA Broadcast — free, if you have the GPU
Free · no affiliate relationship
NVIDIA Broadcast is a free app that turns an RTX-equipped Windows PC into a processed audio/video device — same virtual-microphone architecture as Krisp, different silicon doing the work. For this site's lane the relevant feature is Room Echo Removal, which strips the reverberant ring from your voice; it ships alongside Noise Removal and the newer Studio Voice mode. Requirements, verified 2026-09-20 from NVIDIA's product page: a GeForce RTX 2060, Quadro RTX 3000, TITAN RTX or better, Windows 10/11 64-bit, and 8 GB of RAM.
The honest framing: if you already own the GPU, this costs nothing and does the room-echo half well, with processing running on tensor cores instead of your CPU. What it isn't is a cross-platform answer — there's no macOS build, no non-RTX path, and its focus is your outgoing voice quality rather than the full two-sided echo problem. Gamers and streamers with RTX cards often have it installed already and just haven't enabled the audio effects. We have no affiliate relationship with NVIDIA; the link earns us nothing: nvidia.com — Broadcast app.
Side-by-side
| Option | Cost (verified 2026-09-20) | Platforms | Echo coverage | Best for |
|---|---|---|---|---|
| Built-in app / browser AEC | Free, already installed | Everywhere | Acoustic echo, common-case setups | First attempt for everyone; laptops at sane volume |
| Krisp | Free 60 min/day; from $8/user/mo annual ($16 monthly) | Windows, macOS (desktop only) | Acoustic Echo Cancellation + Room Echo Cancellation, system-wide across all apps | External-speaker setups, echoey rooms, people living in 3+ calling apps |
| NVIDIA Broadcast | Free | Windows with RTX 2060+ GPU | Room Echo Removal (outgoing voice) | RTX owners who sound hollow; zero-budget fix on gaming rigs |
What we left out, and why
- Hardware — headsets, speakerphones, DSP mics. Hardware can absolutely end an echo problem, and honesty requires saying that wearing headphones breaks the loop outright. But this is a software site; we don't review or link gear, and you don't need any to fix echo.
- Post-production echo removal. Tools that de-reverb a recording after the fact solve a different problem on a different timeline. Out of scope, as noted up top.
- Niche telephony/PBX echo cancellers. There's a small ecosystem of tools aimed at call platforms and PBX operators rather than people on calls. If you run a phone system you already have vendors for this; nothing there beats the three options above for an individual.
- OS-level mic modes. macOS ships a Voice Isolation mic mode and Windows machines vary wildly in driver-level "enhancements." They're worth knowing about, but neither is a dependable cross-app echo answer, and per-OS deep dives aren't this page's job.
How to choose, by situation
- Echo only in one app, everything default? Use that app's own AEC — check it hasn't been disabled, and check nothing rerouted your audio. Cost: $0.
- You're told there's echo whenever YOU join, on external or loud speakers? Krisp's Acoustic Echo Cancellation is aimed at exactly this. Burn the free hour on your next real call and see.
- People say you sound hollow / "in a bathroom"? RTX GPU on Windows: NVIDIA Broadcast, free. Otherwise: Krisp's room echo handling, same free-hour test.
- Echo follows you across three different calling apps? That's the system-wide case — a virtual-mic tool (Krisp) beats fixing three apps separately.
- The echo is someone else's device? No purchase helps. Find them, then send them here.
FAQ
Can echo cancellation software fix echo caused by another participant?
Mostly no. AEC running on your machine stops your speakers from re-entering your mic. If someone else's device is looping, the fix happens on their machine. Some tools reduce inbound mess, but the reliable move is identifying the source — our group-call guide takes about two minutes.
Is Krisp's free tier really enough to evaluate it?
For echo, yes. Echo either stops or it doesn't — you'll know within one call. The 60-minute daily allowance (verified 2026-09-20) means a long meeting will hit the ceiling, but evaluation doesn't need a long meeting.
Why not just wear headphones?
Headphones break the speaker-to-mic loop completely, and if you're comfortable in them, do that — it's free. This guide exists because plenty of people can't or won't wear them all day; the speakers-only playbook is written for exactly that.
Will two echo cancellers conflict — say, Krisp plus my meeting app's AEC?
Generally they coexist: the dedicated tool cancels first and the app's AEC finds little left to do. Doubling up on noise processing can over-thin a voice, so vendors suggest letting one layer lead. If your voice sounds robotic with both active, turn the app-level processing down and keep the system-wide layer.
Does any of this help with the hollow sound on my recordings?
Live processing improves what gets recorded going forward (process first, record the processed feed). Cleaning up an existing file is post-production — a different toolchain, and out of this site's scope.