AI models: Noise Cancellation

Introduction

The following document lists Sanas' Noise Cancellation (NC) model releases. Versions are listed newest-first. Each entry summarizes what changed and who it affects.

Sanas NC uses independently trained AI models for each audio direction — outbound (agent-side) and inbound (customer-side) — because each faces distinct acoustic challenges. Outbound models handle near-field speech in noisy environments, while inbound models handle far-field speech from unpredictable sources like speakerphones or car audio.

For desktop application releases that bundle these models, see Product updates. For detailed benchmarking methodology and performance data, see Sanas science blogs.

NC2.6

Whats changed

• First inbound NC model, enabling Omnidirectional Noise Cancellation across all call participants.

• Purpose-built for far-field and variable-quality audio sources (speakerphones, car audio, mobile mics), where distant speech is the signal to preserve rather than suppress.

• Handles unpredictable ambient environments (background TVs, traffic, other people) through an AI model trained specifically for inbound scenarios.

NC1.2

What changed

• Cancels background noise up to 80–90 dBA, significant improvement over the 60–70 dBA range typical of other solutions. Built for high-density contact center floors where adjacent agents are often audible on calls.

• Goes beyond environmental noise to cancel background speech using spatial and acoustic cues such as distance-to-mic, preserving only the foreground (main) speaker.

• Supports over 8,000 noise profiles covering both contact center and work-from-home environments, with no hardware installation required.

Support