How to Add Real-Time Noise Cancellation to a Voice Agent (Python Quickstart)

Background noise is one of the fastest ways to break a voice agent. A dog barking in the background, a passing car, an open-plan office: any of these can push a well-tuned voice pipeline into misheard intents, garbled transcripts, or a frustrated caller repeating themselves. The fix does not have to be complicated. With a first-party SDK, real-time noise cancellation can be added to an existing voice agent stack in a small number of lines of code, without re-architecting the pipeline.

Why SDK-level integration matters

Noise cancellation can be implemented on the client device, on a media server, or inside the voice agent’s own audio pipeline before it reaches speech-to-text. Doing it at the SDK level gives you the most control over latency and avoids an extra network hop.

A basic Python integration

Intercept each incoming audio frame, process it with the Eigen SDK, then pass the cleaned frame downstream to your existing STT engine. The same pattern applies in Node, without requiring separate infrastructure or routing changes.

What real-time needs to mean in practice

Eigen is built for real-time processing with latency under 10 milliseconds, which keeps the added processing step effectively invisible inside a live call. Its language-agnostic processing supports multilingual call centers without swapping models per language.

Getting started

Run a controlled test with a sample of your noisiest recorded calls, compare transcription and intent-detection accuracy before and after cleanup, then validate the integration during Eigen’s 14-day free trial.