---
name: voice-clone-safety-check
description: Run a set of consent + provenance checks before a voice cloning job is accepted.
title: Voice Clone Safety Check
category: voice-audio
difficulty: advanced
license: MPL-2.0
author: admin
source_url: "https://github.com/coqui-ai/TTS"
icon: 🛡️
input: audio
output: structured-json
phase: pre
domain: ops
tags: voice-cloning,consent-verification,audio-fingerprinting,asr,safety-gate,blocklist,pre-flight-check,policy-enforcement,audit-logging,abuse-prevention,workflow-gating,voice-synthesis
best_for:
  - voice cloning pipelines
  - audio generation deployment
  - consent verification workflows
  - regulatory compliance gates
---

## Description

A pre-flight safety skill for voice-cloning workflows. Verifies that the source audio includes an on-clip consent phrase, checks it against a blocklist of public figures / protected persons, and forces a logged consent record before any model training or inference runs.

## Why it works

Voice cloning without consent is the #1 abuse vector for audio generation tools, and the policy / liability risk is the main blocker on commercial deployment. Gating the *pipeline entry point* (not the model) means every downstream model inherits the same check — and the check is auditable. Embedding the consent phrase detection in the audio itself (rather than a checkbox) materially raises the bar for bad-faith use.

## How it works

1) Transcribe source audio with a fast ASR (Whisper tiny / distil-whisper). 2) Fuzzy-match the transcript against a required consent phrase template ('My name is X and I consent to my voice being used for Y on DATE'). 3) Hash the audio fingerprint and check against a blocklist (hashed public-figure samples + customer-supplied blocklist). 4) If any check fails, the skill returns `{allowed: false, reason: ...}` and halts the workflow. 5) If it passes, it writes a signed consent record (actor_id, audio_hash, transcript, timestamp) to an append-only log and returns `{allowed: true, consent_id}`.