AI Guides

How to remove background noise from microphone audio using AI tools

by RebootTips

The Persistent Noise Problem

The other day, I was trying to patch into a critical vendor call from my home office. My old workstation fan decided to ramp up to jet-engine levels right as I needed to explain a network architecture diagram. The client kept asking me to repeat myself, and I could practically hear their patience wearing thin. I’ve used noise gates before, sure, but a constant, dynamic fan noise combined with the occasional dog bark from next door? Traditional methods just made my voice sound choppy, or they let too much through. That’s when I finally decided to put some of these ‘AI noise removal’ tools through their paces for real-time communication.

Why AI Is Different

Look, for simple, constant hums, a basic noise gate in OBS or even some headset drivers can do a decent job. But the world isn’t static. You get intermittent sounds, variable fan speeds, keyboard clatter, kids yelling. A static noise gate either cuts too much out, making you sound like you’re talking through a walkie-talkie, or it doesn’t cut enough, letting all the nuanced chaos bleed through. The promise of AI here isn’t just about filtering; it’s about intelligently differentiating between human speech and everything else, adapting on the fly. It’s a different league entirely when you’re dealing with live audio and you need clarity without sounding artificial.

Setting Up AI Noise Reduction for Your Microphone

I focused on two main approaches I’ve found reliable for live communication: dedicated software solutions and post-processing tools. For real-time calls, you need something that integrates directly as a virtual microphone.

Method 1: Real-time AI with Dedicated Applications (e.g., NVIDIA Broadcast, Krisp)

This is my preferred method for live calls and streaming. These tools create a virtual microphone device that other applications can then select.

  1. Choose Your Tool:
    • NVIDIA Broadcast: If you have a supported NVIDIA RTX graphics card (20-series or newer), this is often the most performant option as it leverages the GPU.
    • Krisp: A software-only solution that works on most systems. It offers excellent noise suppression and can also remove noise from incoming audio, which is a nice bonus for noisy colleagues. There’s a free tier for a certain amount of minutes per week, or a paid subscription for unlimited use.
    • Other options exist, but these are the two I’ve had the most consistent success with.
  2. Download and Install: Grab the installer from the official website (e.g., nvidia.com/broadcast or krisp.ai) and follow the prompts. Restart your machine if requested.
  3. Configure the AI Application:
    • For NVIDIA Broadcast:
      1. Open the NVIDIA Broadcast application.
      2. Navigate to the Microphone tab.
      3. Under Source Microphone, select your physical microphone (e.g., HyperX QuadCast S or Integrated Microphone (Realtek)).
      4. Under Effects, click the plus sign and choose Noise Removal.
      5. Adjust the Strength slider. I usually start around 60-70% and tweak it based on the current background noise. Too high, and your voice can start to sound a bit “processed” or robotic.
    • For Krisp:
      1. Launch the Krisp application (it often lives in your system tray).
      2. In the Krisp window, select your actual microphone under Microphone.
      3. Ensure Remove Noise is toggled On for the microphone.
      4. There’s typically a simple toggle, not a strength slider, which simplifies things.
  4. Integrate with Your Communication App: This is the crucial step.

    In your conferencing software (Zoom, Teams, Discord, OBS, etc.), you need to tell it to use the *virtual* AI microphone.

    1. Open the settings in your chosen application.
    2. Go to the Audio or Voice & Video section.
    3. For your Microphone Input Device (or similar setting), select:
      • NVIDIA Broadcast (NVIDIA RTX Voice) if using NVIDIA Broadcast.
      • Krisp Microphone if using Krisp.
    4. Perform a quick mic test within the app if available, or just speak and listen to how it sounds.

Method 2: Post-Processing with Audacity (or Online Tools)

This method is for recorded audio that you can edit later. It’s not suitable for live calls but gives you more control.

  1. Record Your Audio: Use your preferred recording software (Audacity, OBS Studio, etc.) to capture your audio with your physical microphone.
  2. Import into Audacity: Open Audacity and drag your audio file into it.
  3. Use Noise Reduction: Audacity has a built-in noise reduction effect, which while not strictly “AI” in the modern sense, has become quite effective over the years. Some users also integrate VST plugins that leverage AI.
    1. Select a section of your audio that contains only background noise (no speech).
    2. Go to Effect > Noise Reduction.
    3. Click Get Noise Profile.
    4. Select your entire audio track.
    5. Go back to Effect > Noise Reduction.
    6. Adjust the sliders for Noise Reduction (dB), Sensitivity, and Frequency Smoothing. I usually start with the default values or slightly increase reduction and sensitivity, then preview.
    7. Click OK when satisfied.
  4. Consider Online AI Tools: For quick cleanups without dedicated software, services like Adobe Podcast’s Enhance Speech feature (free for limited use) do a remarkably good job. You simply upload your audio file, and it processes it using AI.

Typical issues

I’ve run into a few common snags over the years with these sorts of setups.

  • The biggest one, and frankly, my own silly mistake the first time: I’d configure NVIDIA Broadcast perfectly, see the noise disappearing in its own monitoring, then jump into Zoom only to find the noise was still there. I’d forgotten to actually *select* NVIDIA Broadcast (NVIDIA RTX Voice) as the input device within Zoom’s audio settings. It’s not enough to set it up in the standalone app; you have to tell every other application to use the virtual microphone. Always check your application’s audio input device.
  • Over-processing: Setting the noise reduction too high, especially on tools with a strength slider, can make your voice sound unnatural, like you’re talking through a tin can, or even cause strange artifacts. Less is often more; aim for clarity, not absolute silence.
  • Resource Usage: AI noise reduction, particularly GPU-accelerated ones, uses system resources. On older or lower-spec machines, you might notice a slight performance hit or increased latency. Monitor your CPU/GPU usage if you encounter stutters or delays.
  • Conflicting Software: If you have multiple noise reduction tools or audio enhancement suites running (e.g., your headset’s software, Windows’ built-in enhancements, and a third-party AI tool), they can conflict. I always recommend disabling all other processing in your sound card drivers or Windows sound settings when using a dedicated AI solution.
  • Driver Issues: Ensure your audio drivers (for your physical mic) and GPU drivers (for NVIDIA Broadcast) are up to date. Outdated drivers can cause unexpected behavior or crashes.

When configured correctly, these AI tools provide a level of audio clarity that traditional methods just can’t match, making online communication significantly less frustrating for everyone involved.

Related Posts