Real-time microphone noise removal

Build Real-Time Microphone Noise Removal with On-Device AI Speech Enhancement

Remove background noise from any microphone audio stream in real time. Turns any microphone into a noise cancelling microphone using on-device AI, with no cloud processing of the audio.

Products used

Koala Noise Suppression

Platforms supported

AndroidiOSLinuxmacOSWindowsChromeEdgeFirefoxSafariRaspberry Pi

Start Free

Loved by developers, trusted by enterprises

How real-time microphone noise removal works

Clean audio from any microphone and isolate the voice from background noise using on-device AI

Koala Noise Suppression processes raw microphone audio frame by frame and outputs a clean signal with background noise removed in real time. Keyboard clicks, fan noise, HVAC hum, street traffic, crosstalk, and ambient chatter are suppressed while preserving speech clarity. Koala runs entirely on-device with no cloud round-trip, so the audio stream is never transmitted externally. The output is a clean PCM stream ready to be piped into a call, a transcription engine, a voice assistant, a recording, or any downstream application.

Why Koala Noise Suppression?

Up to 17.3x more effective than RNNoise at restoring speech intelligibility

17.3×

More effective than RNNoise at 0 dB SNR

5.4×

More effective than RNNoise at 5 dB SNR

4.3×

More effective than RNNoise on average

Koala Noise Suppression removes background noise from audio streams in real time using a lightweight model that runs on-device. The open-source Noise Suppression Benchmark, using the Microsoft DNS Challenge dataset and STOI distance to clean speech as the metric, shows Koala is up to 17.3x more effective than Mozilla's RNNoise at restoring speech intelligibility in the noisiest conditions. Across all noise levels tested, RNNoise reduces STOI distance by a small fraction while Koala cuts it by half or more. No audio is transmitted to Picovoice or any third-party server.

STOI Distance to Clean Speech

Lower is better

Original0.0848

RNNoise0.0748

Koala0.0415

* Benchmarked on the Interspeech 2020 Microsoft DNS Challenge Dataset using the first installment.

STOI Distance to Clean Speech at 0 dB

Lower is better

Original0.232

RNNoise0.226

Koala0.128

* Benchmarked on the Interspeech 2020 Microsoft DNS Challenge Dataset using audio files with 0 dB SNR, where noise and speech are equally loud.

Real-time microphone noise removal use cases

From VoIP calls to field recordings: clean audio streams for any voice application

Calls & Conferencing

Noise-free calls for call centers and VoIP applications

Call center platforms and VoIP providers can integrate Koala to remove agent-side noise (keyboard typing, headset rustling, open-floor crosstalk) and caller-side noise (traffic, wind, public spaces) in real time.

Transcription

Better transcription accuracy from noisy microphones

Cleaning up the microphone input before passing it to a speech-to-text engine improves accuracy and lowers word error rate in noisy environments, such as factories, warehouses, vehicles, and open-plan offices.

Broadcasting

Clean audio capture for broadcasting and streaming

Journalists, event organizers, and content creators can capture clean voice recordings in noisy environments in real time without directional microphones or post-production cleanup.

Meetings

Clean audio for every meeting participant

Meeting platforms, webinar tools, and conferencing SDKs can integrate Koala to remove background noise from each participant's microphone feed before mixing, whether they dial in from their homes, coffee shops, or airport lounges.

Get started

Real-time microphone noise removal: Python code example

A complete working recipe in Python. Open-source on GitHub. Runs 100% on-device.

recipe · real-time-microphone-noise-removal

Difficulty

Beginner

Runtime

100% on-device

Language

Python

Platforms supported

AndroidiOSLinuxmacOSWindowsChromeEdgeFirefoxSafariRaspberry Pi

Prerequisites

Picovoice AccessKey from Picovoice Console and GitHub Repo Clone.

Usage

These instructions assume your current working directory is recipes/call-assist/python.

1

Create a virtual environment

Isolate the recipe's dependencies from your system Python.

2

Activate the virtual environment

Activation makes pip install into .venv instead of system Python.

Linux, macOS, or Raspberry Pi

Windows

3

Install dependencies

Install the Koala Noise Suppression Python SDK and PvRecorder for microphone input.

4

Run the noise removal

Pass your AccessKey. Speak into the microphone and press Ctrl+C to stop. The recipe saves two WAV files: the raw microphone recording and the noise-removed version.

Have questions or looking for implementations in other languages? Visit the GitHub pico-cookbook Real-Time Microphone Noise Removal Recipe, where you can find the open-source demo code and create an issue for demo-related technical questions.

On-device AI cookbook examples

More recipes from picoCookbook

Frequently asked questions

FAQ

+

What is real-time microphone noise removal?

Real-time microphone noise removal is the process of removing background noise from a live audio stream as it is captured, without waiting for the recording to finish. Koala Noise Suppression processes audio frame by frame and outputs clean audio with minimal latency, suitable for live calls, voice commands, transcription, and recording.

+

What types of noise does this microphone noise removal demo remove?

This microphone noise removal demo, powered by Koala Noise Suppression, removes both stationary noise (fan hum, HVAC, electrical buzz, ambient room noise) and non-stationary noise (keyboard clicks, door slams, traffic, crosstalk, paper rustling). It preserves speech clarity while suppressing non-speech audio.

+

How is this demo different from a noise gate?

A noise gate mutes audio below a volume threshold, cutting silence but leaving noise intact whenever someone is speaking. Koala Noise Suppression separates speech from noise within the same audio frame, removing background noise even while the speaker is talking. This produces cleaner results in environments with continuous background noise.

+

How is this microphone noise removal demo different from Krisp or RTX Voice?

Krisp is a subscription SaaS product that processes audio through its own software layer. RTX Voice requires an NVIDIA GPU. Koala Noise Suppression is a licensable SDK that runs on any supported platform (mobile, desktop, embedded, web, Raspberry Pi) with no GPU requirement, no subscription, and no audio sent to a third-party server. You integrate it directly into your application.

+

Does this noise removal demo work offline?

Yes. Koala Noise Suppression runs entirely on-device. No audio is transmitted to Picovoice or any external server. An internet connection is only needed once to validate your AccessKey with Picovoice license servers.

+

Can I use this microphone noise removal demo with a speech-to-text engine?

Yes. Pipe the clean PCM output from Koala Noise Suppression into any speech-to-text engine. Koala pairs directly with any speech-to-text APIs and SDKs. Removing noise before transcription improves word error rate in noisy environments.

+

How can I get technical support?

Visit the GitHub pico-cookbook Real-Time Microphone Noise Removal Recipe, where you can find the open-source demo code and create an issue for demo-related technical questions.