How biometric voice scanning works
Every VoiceSeal scan begins with your enrolled voice fingerprint — a mathematical representation of your unique vocal characteristics generated by our ECAPA-TDNN (Emphasized Channel Attention, Propagation and Aggregation in Time Delay Neural Network) model. This is the same class of model used by enterprise speaker verification systems.
When we scan a platform, we download audio samples from that platform — voice previews, published recordings, or synthesized content — and run the same ECAPA-TDNN model to generate an embedding for that audio. We then compute the cosine similarity between the two embeddings: your enrolled fingerprint and the audio found on the platform.
Cosine similarity produces a score between 0 and 1. Same speaker: typically above 0.92. Different speakers: typically below 0.75. We flag matches at 0.88 or above — a threshold calibrated to eliminate cross-gender and cross-accent false positives while still catching genuine clones. This threshold is higher than most academic benchmarks precisely because we're dealing with voice artists' livelihoods, not a research dataset.
What scanning does not do: VoiceSeal scans compare your voice against audio that has been published on a platform — voice library entries, public previews, or posted content. We do not and cannot intercept synthesis requests made in real time. The pre-synthesis check API (used by platforms that have integrated VoiceSeal directly) is the mechanism that prevents unauthorized synthesis before it happens. Scanning catches what has already been published.
TTS platform scanning — live API accuracy
For TTS platforms that provide API access to their voice libraries, our methodology is direct and highly accurate. We authenticate with the platform's API, retrieve a list of publicly available voices with audio preview URLs, download each preview, and run real biometric comparison against your enrolled fingerprint.
A note on what "accuracy" means here: our 95%+ figure refers to the likelihood that a genuine clone in the platform's voice library will be detected when we scan. It does not mean every scan will find a match — if your voice has not been cloned and published on ElevenLabs, the scan correctly returns clean. A correct negative is not a miss.
The one gap in TTS scanning is private voices. Most TTS platforms allow users to create voice clones that are kept private — not listed in the public library. Our API-based scanning cannot access private voices. We are actively working with platforms to explore consent-based mechanisms that would allow verification of private voice use against the registered owner's fingerprint.
Social media scanning — what we can and can't do
Social media scanning operates on a fundamentally different model. Unlike TTS platforms, social networks do not provide structured voice libraries. Instead, we use keyword-based search and audio extraction to find and analyze public content that may contain cloned voices.
YouTube and Podcasts — most reliable social scanning
We search public content using your voice name and related search terms, extract the audio, and run real ECAPA-TDNN biometric comparison. When we find the right content, accuracy is 85–90%. The primary limitation is coverage: a clone posted under a generic title with no reference to your name or brand may never surface in our search results. This is a discovery problem, not a detection problem — our model is accurate, but we can only analyze content we can find.
Instagram and TikTok — intermittent reliability
Both platforms actively limit automated audio extraction. Our scanning works using yt-dlp and related tooling, but these platforms regularly update their access controls. When extraction succeeds, biometric accuracy is ~85%. However, extraction may fail silently on any given scan — returning no results not because your voice is clean, but because the platform blocked access at that moment. We are transparent about this in scan results and recommend treating a clean Instagram or TikTok result as "no issues found during this scan window" rather than a definitive clearance.
Social scanning is an early-warning system, not a comprehensive monitor. It catches obvious, searchable cases — a clone posted under your name, a viral video using a recognizable version of your voice. It is not designed to catch a bad actor who posts a clone under a generic title with no identifying information.
For voice artists whose work is widely recognized, we recommend supplementing VoiceSeal's automated scanning with periodic manual searches and community reporting. If you encounter a suspected clone that our scanner missed, you can submit it directly from the Detection Alerts section of your dashboard.
Full platform status table
Current integration status for every platform VoiceSeal scans, as of March 2026.
| Platform | Type | Integration | Detection Method | Accuracy |
|---|---|---|---|---|
| ElevenLabs | TTS | Live API | Real audio comparison | ~95%+ |
| PlayHT | TTS | Live API | Real audio comparison | ~95%+ |
| Murf.ai | TTS | Live API | Real audio comparison | ~95%+ |
| WellSaid Labs | TTS | Live API | Real audio comparison | ~95%+ |
| Resemble AI | TTS | Live API | Real audio comparison | ~90%+ |
| Speechify | TTS | Live API | Real audio comparison | ~90%+ |
| YouTube | Social | Live | Keyword search + real audio comparison | ~85–90% |
| Podcasts | Social | Live | RSS directory + real audio comparison | ~80–85% |
| Social | Partial | Real audio comparison when accessible | ~85% / ~50–60% reliable | |
| TikTok | Social | Partial | Real audio comparison when accessible | ~85% / ~50–60% reliable |
| LOVO AI (Genny) | TTS | Estimated | Platform presence estimation only | Not applicable |
| Suno.ai | Music / AI Audio | Outreach Pending | API access requested — not yet integrated | Pending integration |
| Coqui TTS | TTS / Open Source | Estimated | Platform presence estimation only | Not applicable |
| Replica Studios | TTS | Shut Down | Removed — platform closed 2026 | — |
Our outreach mission — pushing platforms to comply
We believe every AI voice platform has a responsibility to verify that voices in their library were enrolled with genuine consent — and to allow rights holders to verify that their voice is not being used without authorization.
VoiceSeal is actively reaching out to every major TTS platform to request API access that enables real-time voice identity verification. We are not just asking for permission to scan — we are offering to become the neutral, independent verification layer that protects both creators and platforms.
When a TTS platform integrates the VoiceSeal pre-synthesis check API, they demonstrate a genuine commitment to creator consent. Their users can synthesize responsibly. Creators can license their voices with confidence. And platforms protect themselves from legal exposure under BIPA, the NO FAKES Act, and emerging EU AI Act obligations.
Our outreach posture to platforms is straightforward: integrating VoiceSeal is not an admission that cloning has occurred on your platform — it is a public signal that you take creator rights seriously. The platforms that engage early will be recognized as leaders. Those that don't will face increasing scrutiny from regulators, unions, and the voice artist community.
What compliance looks like
A platform that integrates VoiceSeal can:
- Run a pre-synthesis check before generating audio from any enrolled voice
- Block unauthorized synthesis requests automatically
- Verify that a voice was enrolled with proper biometric consent
- Issue license tokens that authorize legitimate use
- Provide audit logs that satisfy legal discovery requirements
As platforms comply, our scanning model becomes more accurate. When a platform integrates our pre-synthesis API, we gain access to verify voices at the point of synthesis — not just after-the-fact via public library scanning. This is the difference between catching a clone after it's been published and preventing it from being created at all.
Current outreach status
We are in active or planned outreach with the following platforms and organizations. This list will be updated as relationships develop:
- LOVO AI (Genny) — API access requested. Pending response.
- Suno.ai — AI music generation platform increasingly used with voice-like audio. Outreach initiated. Suno's model can produce audio that closely resembles real voices in musical contexts — an emerging vector for voice identity misuse that we are actively working to cover.
- Coqui TTS — Open-source project. We are working directly with community maintainers on a voluntary verification model.
- SAG-AFTRA — Discussions ongoing regarding VoiceSeal as a recommended verification tool for the Interactive Media Agreement framework. Studio contracts expire June 2026 — a critical window for adoption.
- NAVA (National Association of Voice Artists) — We are in an active testing phase with NAVA-connected voice artists. NAVA has not concluded a formal partnership — that conversation is pending the outcome of this testing period, after which we hope to engage in partnership discussions. Feedback from participating artists is directly shaping our detection model and product roadmap.
If you represent a platform, union, or organization and want to discuss integration or partnership, use the form below. We review all submissions personally.
What's coming next
- LOVO AI live integration — pending API access. Converts from estimated to real audio comparison.
- Continuous social monitoring — moving from periodic scans to near-real-time monitoring for YouTube and podcast platforms for Creator Pro subscribers.
- Private voice verification — working with TTS platforms to build consent-based mechanisms that allow registered owners to verify their voice is not being used in private clones.
- Platform transparency score — a public rating for each TTS platform based on their cooperation with voice identity verification. Updated quarterly.
- EU AI Act Article 50 compliance — machine-readable synthetic audio marking required by August 2, 2026. VoiceSeal watermarking supports this standard.
Every platform that integrates VoiceSeal's pre-synthesis check API will be listed publicly on this page as a verified compliant partner. We believe transparency about who is doing the right thing — and who isn't — is itself a form of accountability.