Skip to navigation

This page covers Helm values for the sender and receiver components of the nvidia-active-speaker-detection-h4m-sample chart. NIM service keys within the demo chart live under nvidia-active-speaker-detection-h4m-service.<key>. For more information, refere to Common Helm Configuration and Pipeline Configuration.


Sender

ConfigurationHelm Key (sender.*)Comment
Enable sendersender.enabledAllowed: Boolean; default: true. Two-phase bring-up (receiver + NIM service first, then sender) recommended.
Naming / scalesender.appName, sender.replicasDefaults: nvidia-active-speaker-detection-sender-nmos, 1.
Nodesender.nodeSelector.hostnameDefault: example-gpu-node; set to a node name that exists in your cluster (such as from kubectl get nodes).
Imagesender.image.repository, sender.image.tag, sender.image.secretDefaults: nvcr.io/nim/nvidia/active-speaker-detection-h4m-sample-sender, 1.0.0, ngc-api-key.
Input assetsender.inputAssets.*Keys: videoFile, numOfAudioTrack, videoFramerateNum, videoFramerateDen, videoWidth, videoHeight; defaults: /workspace/assets/sample_asset.ts, 2, 30, 1, 1920, 1080.
SMPTE ST 2110 portssender.st2110Ports.*Keys: senderRemoteIp, video, audio; defaults: 234.5.8.26, 7005, 7006.
NMOSsender.nmos.*Keys: enabled, hostname, description, label; defaults: true, asd-sender.local, NVIDIA Active Speaker Detection Sender, Nvidia-ASD-Sender. Same semantics as NIM service nmos.*; refer to Common Helm Configuration.
Multus / secondary networksender.network.nameDefault: media-a-tx-net; pod secondary network attachment (such as high-speed media fabric).

Input Asset Requirements

The sender expects an MPEG Transport Stream (.ts) file with the following specifications:

  • Video stream: H.264, with one or more visible speakers.
  • Audio streams: Opus, 48 kHz, mono; one track per speaker.

The sender container includes a bundled sample file at /workspace/assets/sample_asset.ts (1080p, 30 fps, 2 audio tracks).

An audio stream must contain only a single speaker’s audio. During silence, samples must be zero. If background noise is present, enable audio thresholding via useAudioThresholdToDetectActiveAudioStream and audioThresholdDb on the NIM service. For details, refer to Pipeline Configuration.


Receiver

ConfigurationHelm Key (receiver.*)Comment
Enable receiverreceiver.enableddefault: true.
Naming / scalereceiver.appName, receiver.replicasDefaults: nvidia-active-speaker-detection-receiver-nmos, 1.
Nodereceiver.nodeSelector.hostnameDefault example-gpu-node; set to a node name that exists in your cluster (for example, from kubectl get nodes).
Imagereceiver.image.*Defaults: repository nvcr.io/nim/nvidia/active-speaker-detection-h4m-sample-receiver, tag 1.0.0, secret ngc-api-key.
Receive modereceiver.receiverTypeAllowed: ancillary or video; default: ancillary (see next row).
Ancillary paramsreceiver.ancillaryInputParams.*Keys: hostIp, hostPort, framerateNum, framerateDen; defaults: 234.5.8.27, 7006, 30, 1. Used in ancillary mode.
Video paramsreceiver.videoInputParams.*Keys: videoReceiverRemoteIp, videoReceiverPort, videoWidth, videoHeight, videoFramerateNum, videoFramerateDen; defaults: 234.5.8.26, 7005, 1920, 1080, 30, 1. Used in video mode.
NMOSreceiver.nmos.*Keys: enabled, hostname, description, label; defaults: true, sd-receiver.local, NVIDIA Active Speaker Detection Receiver, Nvidia-ASD-ANC-Receiver.
SRT outputreceiver.srtPort.internal, receiver.srtPort.externalSRT preview stream mapping; defaults: internal 8888, external 30889.
Multus / secondary networkreceiver.network.nameDefault: media-a-tx-net; pod secondary network attachment (such as high-speed media fabric).

Receiver Modes

ModeValueDescription
AncillaryancillaryReceives bounding box metadata via SMPTE ST 2110-40. Use to consume detection results programmatically.
VideovideoReceives output video with bounding box overlays, re-streamed via SRT. Requires testFrameOverlayMode: true on the NIM service.

See Also