Is this page helpful?

Operator Configuration#

This page covers Helm values for the nvidia-active-speaker-detection-h4m-operator chart (controller pod) and fields for the NvidiaActiveSpeakerDetectionMediaFunction custom resource (NIM workload).

Pipeline tuning semantics on the custom resource match the Pipeline Configuration Helm keys.

Operator Helm Chart (nvidia-active-speaker-detection-h4m-operator)#

Values for the operator controller pod (not the NIM deployment):

Configuration	Helm Key	Comment
Operator image	`image.repository`, `image.tag`, `image.pullPolicy`	Defaults: `nvcr.io/nim/nvidia/active-speaker-detection-h4m-operator`, tag `1.0.0`, pull `Always`.
Inline registry credentials (operator pod)	`imageCredentials.*`	Defaults: registry `nvcr.io`, username `$oauthtoken`, password `""`, email `""`; empty password skips chart-created pull secret.
Inline registry credentials (NIM operands)	`mediaFunction.imageCredentials.*`	Same shape as `imageCredentials`; empty password skips the media-function registry secret.
Pull secrets	`imagePullSecrets`	Default: `[{name: ngc-api-key}]` (operator pod image pull).
Default NIM image for operands	`mediaFunction.image`, `mediaFunction.imagePullSecrets`	Defaults: image `nvcr.io/nim/nvidia/active-speaker-detection-h4m-nim:1.0.0`, pull secrets `[{name: ngc-api-key}]`.
Default NIM scheduler	`mediaFunction.schedulerName`	Default: `topo-aware-scheduler`.
Operator replicas	`replicas`	Default: `1`.
Leader election	`leaderElection`	Default: `true`.
Operator pod resources	`resources.*`	Defaults: limits `500m` CPU / `128Mi` memory; requests `10m` CPU / `64Mi` memory.
Metrics	`metrics.enabled`, `metrics.port`	Default: `true`, port `8443`.
ServiceAccount	`serviceAccount.create`, `serviceAccount.name`	Default: create `true`, name `""`.

Important

The media function image is configured when the operator is installed, so NvidiaActiveSpeakerDetectionMediaFunction instances managed by an operator use the same validated operator and media-function image combination. This approach ensures the operator and workload are always supported together and avoids compatibility issues.

As a result, CRs managed by an operator share the chart-level mediaFunction.image and mediaFunction.imagePullSecrets configured at install time. To run multiple custom resources with different NIM image tags, deploy multiple operators, each configured with its own mediaFunction.image.

Custom Resource (NvidiaActiveSpeakerDetectionMediaFunction)#

Full YAML example: Example Manifest.

Scheduling and Cluster#

Configuration	CR Path	Comment
Node selector	`spec.nodeSelector`	Kubernetes node labels (for example, `kubernetes.io/hostname`).

Parameters: Network, Resources, Secrets, Security#

Configuration	CR Path	Comment
High-speed networks	`spec.parameters.highSpeedNetwork`	Default name: `media-a-tx-net`; for Multus, list of `{name, ip?}`.
Pod resources	`spec.parameters.resources`	Kubernetes requests and limits for the operand NIM workload.
NGC model download	`spec.parameters.ngcModelDownload`	`secretName`, `secretKey`; default key: `NGC_API_KEY`.
Model cache / logs	`spec.parameters.nimModelCache` / `spec.parameters.nimLogs`	Optional PVC-backed model cache and NIM logs. Subkeys include `path`, `pvc.enabled`, and `pvc.claimName`; refer to Advanced Usage.
Security context	`spec.parameters.securityContext`	Example: `runAsUser: 1000`, capabilities for Rivermax and RTP; refer to Getting Started
Pod security context	`spec.parameters.podSecurityContext`	Example: `fsGroup: 1000`, `seccompProfileType: RuntimeDefault`; applies at the pod level.
NMOS metadata	`spec.parameters.label`, `spec.parameters.description`	NMOS IS-04 registration strings.

Pod Resources (spec.parameters.resources)#

Resource	Requests	Limits
`cpu`	`12`	`12`
`hugepages-2Mi`	`8Gi`	`8Gi`
`memory`	`8Gi`	`8Gi`
`nvidia.com/gpu`	`1`	`1`

Parameters: Pipeline Tuning#

Configuration	CR Path (`spec.parameters.*`)	Environment Variable (on NIM Pod)	Comment
Sync tolerance	`syncTolerance`	`AI4M_ACTIVESPEAKERDETECTION_SYNC_TOLERANCE`	Default: `0.5986`; valid range: `(0, 1)`. Invalid values fall back to default.
Bounding box overlay	`testFrameOverlayMode`	`AI4M_ACTIVESPEAKERDETECTION_TEST_FRAME_OVERLAY_MODE`	`true` or `false`
Output frame buffer size	`outputFrameBufferSize`	`AI4M_ACTIVESPEAKERDETECTION_OUTPUT_FRAME_BUFFER_SIZE`	Default: `30` frames.
Audio silence detection	`useAudioThresholdToDetectActiveAudioStream`	`AI4M_ACTIVESPEAKERDETECTION_USE_AUDIO_THRESHOLD_TO_DETECT_ACTIVE_AUDIO_STREAM`	Default: `false`.
Audio silence threshold	`audioThresholdDb`	`AI4M_ACTIVESPEAKERDETECTION_AUDIO_THRESHOLD_DB`	Default: `-40.0` dB.
Log level	`loggingLevel`	(mapped by operator to NIM env)	`0` FATAL, `1` ERROR, `2` WARN, `3` INFO (default), `4` DEBUG, `5` VERBOSE.

Inputs and Outputs#

Area	CR Path	Comment
Video in	`spec.inputs.video_input`	`control`, `transport`, `format.video`; see subfields in next table.
Audio in	`spec.inputs.audio_inputs` (`audio_input_0` … `audio_input_<n-1>`)	One block per stream; count must match pipeline.
Video out	`spec.outputs.video_output`	Optional; resolution must match input when present.
Ancillary out	`spec.outputs.ancillary_data_output`	Required for SMPTE ST 2110-40; `format.data.media_type` default: `video/smpte291`; payload layout: Ancillary Data Payload.

Video Format Subfields#

These subfields apply to video input and video output.

Field	Example Value	Comment
`media_type`	`video/raw`	Fixed for uncompressed ST 2110-20.
`frame_width`	`1920`	Must match NIM `video.width`.
`frame_height`	`1080`	Must match NIM `video.height`.
`frame_rate`	`"30"`	Frames per second as a string.
`interlace_mode`	`progressive`
`color_sampling`	`YCbCr-4:2:2`
`component_depth`	`10`	Bits per component.
`colorspace`	`BT709`
`transfer_characteristic`	`SDR`

frame_rate must be identical for spec.inputs.video_input.format.video and spec.outputs.video_output.format.video. Invalid values are rejected when the CR is applied. The following values are supported:

Value	Rate
`"24"`	24 fps
`"25"`	25 fps
`"30"`	30 fps
`"30000/1001"`	29.97 fps (fractional)
`"50"`	50 fps
`"60"`	60 fps
`"60000/1001"`	59.94 fps (fractional)

Use a plain integer string for whole rates and a numerator/denominator string for fractional rates. Always quote the value in YAML so fractions are not interpreted as numbers.

Audio Format Subfields#

These subfields apply to each audio_input_n stream.

Field	Example Value	Comment
`sample_rate`	`48000`	Must match NIM `audio.samplingRate`.
`channel_count`	`1`	Mono per stream; must match NIM `audio.numChannels`.
`sample_depth`	`24`	Bits; corresponds to PCM L24.
`media_type`	`audio/L24`	Fixed for ST 2110-30 mono L24.

CR Schema Reference#

With the nvidiaactivespeakerdetectionmediafunctions.nvidia.com CRD installed, inspect field types and OpenAPI details from the cluster:

kubectl explain nvidiaactivespeakerdetectionmediafunctions.nvidia.com --recursive
kubectl explain nvidiaactivespeakerdetectionmediafunctions.nvidia.com.spec
kubectl explain nvidiaactivespeakerdetectionmediafunctions.nvidia.com.spec.inputs.video_input.format.video
kubectl explain nvidiaactivespeakerdetectionmediafunctions.nvidia.com.spec.outputs.ancillary_data_output

Some validation rules might not appear in kubectl explain output but are still enforced when the CR is applied.

Status Conditions (`status.conditions`)#

Status Overview#

The operator reports health using Kubernetes conditions in status.conditions (type, status, reason, message).

A media function is considered healthy and ready when the following are all True:

Provisioned: Deployment and pod are running.
Registered: Successfully registered with NMOS / SDP.
Configured: Required configuration has been applied.

Important Statuses#

Status	What It Means
`Provisioned`	App deployment is running.
`Registered`	Registration succeeded.
`Configured`	Configuration applied successfully.
`Connected`	Connections are established.
`Active`	Reserved for future media-flow status.
`Degraded`	Reserved for future health reporting.

When to Alert#

Alert if any of these stay False:

Provisioned
Registered
Configured

Most important failure reason values:

DeploymentNotAvailable
NmosQueryFailed

Connected False is usually informational or transient and can occur when connections are still being established.

Ignore Active and Degraded for now.

Check Status#

Use the custom resource metadata.name (not a Pod name).

kubectl get nvidiaactivespeakerdetectionmediafunction <cr-name> \
  -o jsonpath='{range .status.conditions[*]}{.type}{"\t"}{.status}{"\t"}{.reason}{"\n"}{end}'