audio-genai-v1

AI-generated audio with recoverable provenance and AudioSeal watermark. Use this recipe when signing audio produced by generative AI systems (text-to-speech, music generation, etc.).

Recipe configuration

id: audio-genai-v1
version: 1
media_type: audio
description: "AI-generated audio with recoverable provenance and AudioSeal watermark"
c2pa:
  enabled: true
  assertions: [ai_generated, generator, model, prompt_hash, created_at]
  kms_mode: managed
watermark:
  enabled: true
  engine: audioseal
  payload:
    type: manifest_pointer
    encoding: compact
soft_binding:
  enabled: true
  methods: [exact_watermark, audio_fingerprint]
anchoring:
  enabled: true
  methods: [opentimestamps]
retention:
  manifest_days: 365
  original_asset_days: 0
  derived_asset_days: 30
verification:
  minimum_confidence: 0.92

Sign an audio file

curl -X POST https://api.verbitas.io/v1/sign \
  -H "Authorization: Bearer vb_test_YOUR_KEY_HERE" \
  -F "[email protected]" \
  -F "recipe=audio-genai-v1" \
  -F 'metadata={"generator":"my-tts-system","model":"v1.0"}' \
  | jq .

What this records

ai_generated — marks the audio as AI-generated
generator — name of the generating system
model — model version used
prompt_hash — SHA-256 of the generation prompt
created_at — ISO 8601 timestamp

Watermark

AudioSeal embeds a localised watermark at PESQ ≥ 3.0 (speech quality). The watermark detects watermarked seconds and survives MP3 re-encoding and format conversion.

Billing

Billed per minute of audio signed.