Voice profile

Voice profiles let Scryon identify the authenticated user among diarized speakers. This is an opt-in feature gated by SCRYON_VOICE_EMBEDDING_ENABLED.

When the feature flag is off, every endpoint below returns 404 feature_not_available.

See Voice embedding for the full feature design.

POST /api/users/me/voice-profile

Create or replace the authenticated user's voice profile.

Request

Content-Type: multipart/form-data with one part:

Part	Required	Notes
file	yes	A clean voice sample of the user. WAV/MP3/M4A. Duration between SCRYON_VOICE_SAMPLE_MIN_SECONDS and SCRYON_VOICE_SAMPLE_MAX_SECONDS.

The client must display consent UI before invoking this endpoint. The current consent version (SCRYON_VOICE_CONSENT_VERSION) is stamped on the row so future product changes can require re-consent.

Response — 201 Created

{
  "status": "ACTIVE",
  "provider": "pyannote",
  "consentVersion": "v1",
  "createdAt": "2026-05-29T13:00:00Z",
  "sampleDurationSeconds": 22
}

Errors

Status	code	Cause
400	voice_sample_too_short	Below SCRYON_VOICE_SAMPLE_MIN_SECONDS.
400	voice_sample_too_long	Above SCRYON_VOICE_SAMPLE_MAX_SECONDS.
413	voice_sample_too_large	Above SCRYON_VOICE_SAMPLE_MAX_SIZE_MB.
415	unsupported_media_type	Unknown audio MIME.
502	voice_provider_error	Provider returned a non-recoverable error.

GET /api/users/me/voice-profile/status

Cheap status check for the client to decide what UI to show.

Response — 200 OK

{
  "status": "ACTIVE",
  "provider": "pyannote",
  "consentVersion": "v1",
  "createdAt": "2026-05-29T13:00:00Z"
}

When the user has never created a profile:

{ "status": "MISSING" }

When the feature is disabled the endpoint returns 404 feature_not_available.

DELETE /api/users/me/voice-profile

Remove the user's voice profile.

Response — 204 No Content

• The row is removed from user_voice_profiles.
• The provider-side embedding (if applicable) is deleted on best-effort.
• Subsequent calls fall back to text-only speaker resolution.

Idempotent — a second call returns 204 even if no profile exists.

Privacy notes

• The voice sample is not retained after the provider produces an embedding. The bytes live only in the request handler.
• The embedding is stored as an opaque provider blob in user_voice_profiles.embedding_json. Scryon does not decode or interpret it.
• The voiceMatchScore exposed on transcripts is a single float in [0, 1] — no biometric data leaks through it.
• Re-uploading replaces the previous profile in place; (user_id) is unique.

See Privacy & security for the full data-handling contract.