Skip to navigation

Overview

View as MarkdownOpen in Claude

NVIDIA Background Noise Removal (BNR) NIM removes unwanted background noise from speech and improves intelligibility in noisy live-production environments. The NIM ingests a single PCM audio stream over SMPTE ST 2110-30, applies real-time denoising, and publishes enhanced audio over SMPTE ST 2110-30 for downstream broadcast equipment.

This NIM is designed for live broadcast and production workflows that ingest and egress uncompressed audio over SMPTE ST 2110 using NVIDIA Rivermax. It is deployed with Helm charts on a Holoscan for Media cluster and can optionally be managed through a Kubernetes operator.

Architecture

The BNR NIM is built on NVIDIA’s software platform:

  • Holoscan for Media: Media platform for SMPTE ST 2110-oriented live pipelines and Kubernetes-native orchestration.
  • NVIDIA Rivermax: GPU-accelerated SMPTE ST 2110 and RTP ingest and egress.
  • CUDA: GPU-accelerated processing.
  • TensorRT: Optimized neural network inference.
  • Triton Inference Server: Local model execution and metrics.

Processing Flow

The NIM performs the following actions in sequence:

  1. Receives uncompressed mono audio via SMPTE ST 2110-30 over the high-speed network interface.
  2. Selects the BNR denoiser variant from model.type (v2-48k or v2-16k).
  3. Runs inference through Triton Inference Server inside the pod.
  4. Applies the configured intensityRatio (from 0.0 to 1.0).
  5. Emits enhanced audio via SMPTE ST 2110-30.

Logical Pipeline

Sender → BNR NIM → Receiver
PieceRole
SenderExample workload: TS file to ST 2110-30 audio (demo chart).
BNR NIMBackground noise removal and, when NMOS is enabled, registration with an external NMOS registry; enhanced ST 2110-30 audio output.
ReceiverConsumes enhanced audio, optionally re-streams via SRT.

Operator variant: Reconciles only the BNR NIM deployment and SDP ConfigMap from an NvidiaBnrMediaFunction custom resource.

Inputs and Outputs

DirectionStream TypeTransportFormatSupported Configurations
InputAudioSMPTE ST 2110-30PCM L24, mono48 kHz with v2-48k; 16 kHz with v2-16k.
OutputAudioSMPTE ST 2110-30PCM L24, monoSame sample rate and channel count as input.

BNR expects a single mono speech stream per input session. For multi-channel mixes, downmix to mono on the sender or instantiate one NIM per speaker channel.

Deployment

BNR NIM supports three deployment patterns:

Deployment OptionDescription
NIM service chartDeploys the BNR NIM as a standalone Kubernetes deployment for integration into a custom media pipeline.
Kubernetes operatorManages BNR NIM workloads through a NvidiaBnrMediaFunction custom resource. Suitable for declarative, production-grade management.
End-to-end demo chartDeploys a complete demo pipeline: sender, BNR NIM service, and receiver. Recommended for initial evaluation.

For installation instructions, refer to Installation.

Supported Operating Modes

ST 2110 Static Mode

Media streams use fixed multicast IP addresses, ports, and session names. Configure the sender output, BNR input/output, and receiver input so the IP and port chain is consistent.

NMOS Mode

Devices and streams are registered and connected using AMWA NMOS IS-04 and IS-05. This mode simplifies setup by moving transport negotiation into the NMOS Connection Manager UI.

See Also