Skip to content

Speaker Services - Overview

Speaker Services provide user-owned speaker identities, enrollment recordings, automatic biometric evidence, verification, and analytics. Verification resources are grouped under /api/v1/speaker-services/speaker-verification; embedding model discovery remains under /api/v1/speaker-services/speaker-embeddings.

Service Purpose
Speakers Identity, lifecycle, and enrollment-based readiness.
Enrollments Model-independent source recordings and enrollment reset.
Voice Prints Automatically generated immutable evidence, optional quality preview, and retention.
Verifications Perform speaker matching and optional sentence verification.
Embedding Models Available embedding models and the current default.
Analytics Verification results by speaker, recorded model, model version, or evidence.

Typical Integration

  1. Create a speaker, initially ACTIVE and NOT_READY.
  2. Submit enrollments until enrollmentSummary.usable reaches minimumRequired (three by default). Usable recordings are READY, retained, and neither deleted nor purged. verificationStatus then becomes READY.
  3. Optionally preview quality. The console shows this preview automatically using the default model as you add samples. Preview performs no persistence and does not gate verification.
  4. Verify. Omit model to follow the current default, or supply an explicit model. The service resolves the model once and automatically obtains matching evidence.
  5. Inspect the returned voicePrintId, requestedModelAlias, modelAlias, resolvedModel, and modelHash. Read the evidence or query analytics as needed.

Model Changes

Changing the default model affects subsequent requests that omit model. Updating a model's underlying version affects subsequent requests using that model. Model selection determines which evidence is compatible with each request.

A completed print is reused when the model version and usable enrollment set match. Otherwise all enrollment embeddings and the voice print are rebuilt and saved together. Requests already in progress retain their resolved model. Each model version has fixed weights.

Retention and Limits

Usable enrollment recordings are preserved for active speakers so voice prints can be generated with updated models. Historical verification references protect biometric evidence from automatic cleanup. Explicit deletion, reset, and purge remain available; explicit purge can remove historical artifacts.

Enrollment and verification quotas, request rate limits, and overall biometric storage limits apply. If required audio is unavailable, re-enrollment is necessary.