Skip to main content
Real-time noise cancellation for human communication
Model: VI_G_NC3.0
Max Sample Rate: 24 kHz
Latency: ~40ms (streaming)
Use Case: Human calls, conferencing, gaming, media

​
Introduction

Sanas Noise Cancellation (NC) provides AI-powered noise cancellation that extracts voices from noisy signals in real-time across any acoustic conditions. With low latency and minimal computing cost, NC delivers clean speech waveforms suitable for human communication.

​
Why Sanas Noise Cancellation?

Real-Time Processing


~40ms latency is suitable for live conversations without noticeable delay

Natural Voice Quality


Preserves voice characteristics and emotional tone while removing noise

Low Compute Cost


Efficient processing enables high concurrency on standard hardware

Versatile Application


Works across various acoustic scenarios and microphone types

​
Model Overview

​
Voice Isolation - General

Preserves: Primary speaker up to ~1m
Removes: Background voices, noise, and chatter
Max Sample Rate: 24 kHz Use cases:
  • General-purpose microphones
  • Voice agents and bots
  • Customer service calls
  • Single-speaker recordings
Configuration: 
audio_params.modelName = "VI_G_NC3.0"
audio_params.sampleRate = 16000  # Or up to 24kHz

​
Audio Samples

Example: The source audio sample was recorded in an outdoor setup with multiple speakers nearby and other ambient noise. Sanas Voice Isolation preserves the main speaker’s voice while eliminating all other voices and ambient noises. Source audio Enhanced audio (Voice Isolation)

​
Setup Guide

1

Configure the model

Set the model name in your audio parameters:
    audio_params.modelName = "VI_G_NC3.0"
    audio_params.sampleRate = 16000  # Or up to 24kHz
2

Initialize the SDK

Follow the complete setup instructions in the Quickstart Guide
3

Start processing

Begin streaming audio for real-time noise cancellation
For complete setup instructions, see the Quickstart Guide.

​
Use Case Examples

Scenario: Agent-customer calls with office background noise
    audio_params.modelName = "VI_G_NC3.0"
    audio_params.sampleRate = 16000
Result: Removes office background while preserving natural call quality and the agent’s voice clarity.
Scenario: Remote worker with home office noise
    audio_params.modelName = "VI_G_NC3.0"
    audio_params.sampleRate = 24000
Result: Eliminates AC noise, keyboard typing, and household sounds while maintaining voice quality.
Scenario: Gamer with keyboard, mouse, and fan noise
    audio_params.modelName = "VI_G_NC3.0"
    audio_params.sampleRate = 24000
Result: Provides maximum clarity by removing all peripheral device noise while preserving voice.
Scenario: General microphone in various environments
    audio_params.modelName = "VI_G_NC3.0"
    audio_params.sampleRate = 16000
Result: Balances noise removal with natural voice preservation for typical microphone distances.

​
Latency & Performance

MetricValueDescription
Streaming Latency~40msPer-chunk processing time
ThroughputReal-time+Processes faster than real-time
ConcurrencyUnlimitedLimited only by your infrastructure

​
Next Steps

Quickstart Guide

Complete setup and initialization instructions

API Reference

Explore the SDK API reference

Multi-Stream Tutorial

Handle multiple concurrent calls
Need help? Contact our support team at support@sanas.ai or raise a ticket.