POST /enroll - VoiceMark

Enrols a user’s voice. Extracts an ECAPA-TDNN 192-dim speaker embedding and stores it keyed by a UUID. Raw audio is deleted immediately after processing.

Request

POST /v1/enroll Content-Type: multipart/form-data

Field	Type	Required	Notes
`audio`	File	Yes	WAV, WebM, MP3 or M4A. Minimum 2 seconds of speech.
`user_id`	string	Yes	Your internal identifier for this user. Not stored after processing.

Response

{
  "voiceprint_id": "3f2a1b4c-e8a9-4f7d-b2c1-9d8e3f1a7b5c",
  "enrolled_at": "2026-11-09T14:32:00Z",
  "quality_score": 0.82,
  "duration_s": 4.3,
  "glyph_params": {
    "hue": 0.61,
    "shapeStretch": 0.44,
    "shellDensity": 0.73,
    "breathPulse": 0.38,
    "fluxMotion": 0.55,
    "scatterAir": 0.21,
    "chaosWarp": 0.18,
    "pulseDepth": 0.47,
    "kurtosisIntensity": 0.62,
    "colorVibrancy": 0.80
  }
}

Field	Description
`voiceprint_id`	Store this — pass to `verify()` to enable identity check
`enrolled_at`	ISO 8601 timestamp
`quality_score`	0–1. Prompt re-record if below 0.4. See quality scoring.
`duration_s`	Duration of usable speech in the clip
`glyph_params`	Render the Glyph VoiceMark immediately

Error codes

Code	Status	Cause
`AUDIO_TOO_SHORT`	400	Clip under 2 seconds
`AUDIO_TOO_QUIET`	400	RMS energy below threshold
`LOW_VOICE_SIGNAL`	400	No recognisable voice identity in audio
`AUDIO_CONVERSION_FAILED`	400	Unsupported format
`UNAUTHORIZED`	401	Missing or invalid API key

⌘I