Deepgram Enhanced Metrics
Beyond the metrics SageMaker itself publishes, the Deepgram container emits its own usage and billing metrics directly into your CloudWatch account. They are written as CloudWatch Embedded Metric Format (EMF) lines on container stdout: SageMaker forwards container output to the endpoint’s CloudWatch log group, and CloudWatch Logs extracts the metrics automatically. No agent, sidecar, or extra IAM permission is required, and the metrics keep flowing even in network-isolated Marketplace deployments.
Two namespaces are published, from two different layers of the container:
| Namespace | Layer | Purpose | Can be disabled? |
|---|---|---|---|
Deepgram/SageMakerInference |
Deepgram’s SageMaker integration | Billing: consumed units per request, with audio duration and character counts | No — always on |
Deepgram/SelfHosted |
Deepgram API server | Usage: per-method, per-tier, and per-feature utilization | Yes — on by default, opt out with an environment variable |
Unlike the Prometheus & OpenTelemetry metrics, these are classic CloudWatch metrics: they appear in aws cloudwatch list-metrics, work with get-metric-statistics, dashboards, and alarms, and need nothing enabled on the endpoint configuration.
All dimensions are low-cardinality and contain no PII — never transcripts, TTS input, or per-request identifiers.
Billing metrics: Deepgram/SageMakerInference
Section titled “Billing metrics: Deepgram/SageMakerInference”One EMF record is emitted per completed request (each streaming session and each pre-recorded or TTS request). These are the same consumed-unit values that drive AWS Marketplace metered billing, so this namespace is the tool for reconciling your AWS bill against actual traffic.
| Metric | Unit | Description |
|---|---|---|
ConsumedUnits |
Count | Billable inference units for the request. Sum over a period is the total billed volume. |
AudioDurationSeconds |
Seconds | Duration of audio processed (speech-to-text requests). |
CharCount |
Count | Characters synthesized (text-to-speech requests). |
BillingLatencyMs |
Milliseconds | Internal latency of the billing pipeline. Useful only for diagnosing billing delays; not a request-latency metric. |
Dimensions are published in three sets — [Category], [Category, Model], and [Category, Model, Transport] — so you can query at any granularity:
| Dimension | Values |
|---|---|
Category |
stt_streaming, stt_batch, tts |
Model |
The deployed model, for example nova-3, flux-general-en, aura-2-thalia-en |
Transport |
ws (streaming), http (pre-recorded / batch TTS) |
Example — total consumed units per hour, per category:
SampleCount is the number of billed requests; Sum is the units consumed.
Usage metrics: Deepgram/SelfHosted
Section titled “Usage metrics: Deepgram/SelfHosted”The Deepgram API server emits a second EMF stream with raw usage broken down by method, model tier, and enabled features — independent of billing. Use it to understand how your endpoint is used: which features are enabled, how much audio each tier processes, and how much of the traffic is streaming versus pre-recorded.
| Metric | Unit | Dimensions | Description |
|---|---|---|---|
AudioMs |
Milliseconds | Deployment, Method |
Audio processed per request (speech-to-text). |
Requests |
Count | Deployment, Method |
Completed requests. |
TtsCharacters |
Count | Deployment, Method |
Characters synthesized (text-to-speech). |
Tokens |
Count | Deployment, Method |
Tokens consumed by intelligence features (summarization, sentiment, and so on). |
VoiceAgentMs |
Milliseconds | Deployment, Method |
Voice-agent connection time. |
TierAudioMs |
Milliseconds | Deployment, Tier |
Audio processed, broken down by model tier (for example nova-3, flux). |
FeatureAudioMs |
Milliseconds | Deployment, Feature |
Audio processed with a given feature enabled. A request with several features enabled attributes its full duration to each. |
FeatureTokens |
Count | Deployment, Feature |
Tokens consumed per intelligence feature. |
Dimension values: Deployment is sagemaker; Method is streaming or sync (pre-recorded); Feature is the request parameter name, for example diarize, smart_format, punctuate, redact, keyterm, interim_results.
Example — how much audio ran with diarization enabled today:
Opting out
Section titled “Opting out”The usage stream is on by default. To disable it, set an environment variable override in the endpoint configuration’s Environment map (see Configure Amazon SageMaker Deployments):
"Environment": {
"DEEPGRAM_API_01": "emf.enabled=false"
}The billing stream (Deepgram/SageMakerInference) cannot be disabled — it is part of the metering pipeline.
Related resources
Section titled “Related resources”- Observability for Amazon SageMaker — SageMaker’s own metrics, logs, and alarms
- Prometheus & OpenTelemetry Metrics — per-GPU and container Prometheus metrics via detailed observability
- Configure Amazon SageMaker Deployments — environment variable overrides
- CloudWatch Embedded Metric Format specification (AWS)