Administration - Sentence Verification Models

Configure sentence verification (sentver) model clients and aliases. Sentver models are Speech Services sentence-verification resources, separate from transcription models, and are currently used by Speaker Services enrollment and verification pipelines.

Base path: /api/v1/speech-services/admin

List Sentver Models

GET /api/v1/speech-services/admin/sentence-verification/models

Requires Authentication - Scopes: admin:sentver-models:write

Returns concrete sentver model client configurations for the tenant.

Register Sentver Model

POST /api/v1/speech-services/admin/sentence-verification/models

Requires Authentication - Scopes: admin:sentver-models:write

Registers a concrete sentver model client. Newly registered models start as INACTIVE; use the status endpoint to make a model available. The name is immutable after creation and remains the concrete model identifier used by aliases and retained records.

Inference calls use round-robin load balancing. grpcTarget accepts a single host:port, dns:///sentver-headless:443 for DNS discovery, an unprefixed list such as server-a:2222,server-b:2222, or the preferred explicit form static:///server-a:2222,server-b:2222. Static lists have fixed membership; use DNS for dynamic membership. DNS targets are refreshed periodically even while another endpoint remains healthy. TLS endpoints must share a certificate identity or use the authority override.

SentverModelAdminRequest

{
  "name": "sentver-he-v2",
  "displayName": "Hebrew Sentence Verification v2",
  "description": "Hebrew sentence verification model client",
  "supportedLocales": ["he-IL"],
  "maxExpectedTextLength": 500,
  "grpcTarget": "dns:///sentver-prod:443",
  "grpcTlsMode": "SYSTEM_TRUST",
  "grpcAuthority": "sentver-prod",
  "grpcDeadlineMs": 10000,
  "grpcMaxInboundMessageBytes": 4194304,
  "grpcMetadata": {
    "x-client": "speaker-services"
  }
}

Update Sentver Model

PUT /api/v1/speech-services/admin/sentence-verification/models/{name}

Requires Authentication - Scopes: admin:sentver-models:write

Replaces editable configuration for an existing concrete sentver model. Required fields must be supplied. Nullable fields may be supplied, omitted, or set to null; omitted nullable fields are cleared.

This endpoint is for configuration fixes such as display text, locale coverage, timeout values, and inference server connection settings. It does not change which concrete model an alias resolves to.

name is immutable and is not accepted in the request body. status is updated only through PATCH /models/{name}/status.

Path Parameters

UpdateSentverModelRequest

{
  "displayName": "Hebrew Sentence Verification v2.1",
  "description": "Hebrew sentence verification model client",
  "supportedLocales": ["he-IL"],
  "grpcTarget": "dns:///sentver-prod-v2:443",
  "grpcTlsMode": "SYSTEM_TRUST",
  "grpcAuthority": "sentver-prod",
  "grpcDeadlineMs": 12000
}

{
  "displayName": "Hebrew Sentence Verification v2.1",
  "supportedLocales": ["he-IL"],
  "grpcTarget": "dns:///sentver-prod-v2:443",
  "grpcTlsMode": "SYSTEM_TRUST",
  "grpcDeadlineMs": 12000
}

Update Sentver Model Status

PATCH /api/v1/speech-services/admin/sentence-verification/models/{name}/status

Requires Authentication - Scopes: admin:sentver-models:write

Updates concrete model availability. Use this endpoint for lifecycle changes such as promoting a model to AVAILABLE, putting it in MAINTENANCE, deprecating it, or making it INACTIVE.

Configuration fields such as displayName, supportedLocales, and gRPC settings are not accepted here. Use PUT /models/{name} for configuration updates.

Path Parameters

UpdateSentverModelStatusRequest

{
  "status": "AVAILABLE",
  "statusMessage": "Promoted after validation"
}

Delete Sentver Model

DELETE /api/v1/speech-services/admin/sentence-verification/models/{name}

Requires Authentication - Scopes: admin:sentver-models:delete

Deletes a concrete model only when no aliases or retained enrollment/verification records reference it.

Alias Management

List Sentver Model Aliases

GET /api/v1/speech-services/admin/sentence-verification/model-aliases

Requires Authentication - Scopes: admin:sentver-model-aliases:write

Returns all sentver aliases for the tenant.

Create or Update Sentver Model Alias

PUT /api/v1/speech-services/admin/sentence-verification/model-aliases/{alias}

Requires Authentication - Scopes: admin:sentver-model-aliases:write

Creates or updates an alias. Exactly one sentver alias per tenant can be marked as default.

SentverModelAliasAdminRequest

Update Sentver Alias Resolution

PUT /api/v1/speech-services/admin/sentence-verification/model-aliases/{alias}/resolve

Requires Authentication - Scopes: admin:sentver-model-aliases:write

Changes the primary resolved model and optional fallback model. Existing enrollment and verification records keep their requested alias and actual resolved model lineage.

Deprecate Sentver Model Alias

PUT /api/v1/speech-services/admin/sentence-verification/model-aliases/{alias}/deprecate

Requires Authentication - Scopes: admin:sentver-model-aliases:write

Marks an alias as deprecated and optionally sets a sunset time. After the sunset time, public requests reject the alias with 410 Gone.

DeprecateSentverAliasRequest

Undeprecate Sentver Model Alias

PUT /api/v1/speech-services/admin/sentence-verification/model-aliases/{alias}/undeprecate

Requires Authentication - Scopes: admin:sentver-model-aliases:write

Removes the deprecated status from a sentver model alias, clearing the deprecation date, sunset date, and deprecation message. The alias becomes fully active again.

Path Parameters

Delete Sentver Model Alias

DELETE /api/v1/speech-services/admin/sentence-verification/model-aliases/{alias}

Requires Authentication - Scopes: admin:sentver-model-aliases:delete

Deletes an alias. Existing retained records remain reproducible because they store both alias and resolved model metadata.

Data Types

ModelThresholdDefaults

Per-mode threshold defaults for a concrete sentver model. All fields are optional and nullable. A null field means "use the global service default" for that mode. Thresholds are floating-point values in the range [0.0, 1.0].

{
  "strictAcceptThreshold": 0.88,
  "strictReviewThreshold": 0.78,
  "balancedAcceptThreshold": 0.72,
  "balancedReviewThreshold": 0.58,
  "lenientAcceptThreshold": 0.58,
  "lenientReviewThreshold": 0.42
}

GrpcTlsMode

SentverModelStatus