capability
verified July 2026
Provenance and privacy, applied at generation.
A cloning engine ships with obligations. This sheet is the three we take on at the engine level — applied at generation, not promised in a PDF — and what each one gives your compliance team to point at.
01
The three commitments
Fig. — Applied at the engine, not appended to a policy
every sample
carries an inaudible watermark, applied at generation on every serving tier
no training
customer text and reference audio never enter a training set
per key
every workload’s traffic auditable on its own key at GET /v1/usage
The watermark is part of synthesis itself — there is no configuration that skips it, on any tier, for any voice.
02
What this gives an audit
Regulated callers — collections agencies, insurers, clinics — end up explaining synthetic audio to someone. The answers are ready before the questions: the audio itself proves it is synthetic (waveform watermark, survives transcoding), the key that generated it is on record (GET /v1/usage), and the reference speaker’s clip went nowhere (no training, fingerprint-cached only for reuse).
03
What we do not claim
- The watermark does not identify the cloned speaker — it marks audio as synthetic, nothing more.
- Consent is enforced as a requirement on you, not detected by us: the speaker in a reference clip has agreed to be cloned.
- Detection tooling evolves; the commitment that does not is where the mark is applied — at generation, on every tier.
04
Notes — an engineer's checklist
01Can I verify the watermark myself?
Bring a sample to the demo call and we run the detector in front of you — the same discipline as the latency readout. Detection is software; the mark itself rides in the waveform.
02Is anything about my traffic used to improve the model?
No — customer text and reference audio never enter a training set. References are fingerprinted and cached solely so reuse of the same clip skips re-cloning.
03What should a compliance review ask a TTS vendor?
Three questions: is synthetic output marked at generation or by policy, does reference audio ever train anything, and can each workload’s traffic be audited separately. This page is our three answers.
See also
Related sheets.
glossary
100%
Audio watermarking
Audio watermarking marks synthesized speech as synthetic. How an inaudible watermark survives where metadata does not, and what a watermark cannot do.
glossary
~1.5 MB
Voice cloning reference audio
The reference is the clip a zero-shot clone is built from: about ten seconds of one clean, consenting speaker, a ~1.5 MB cap, fingerprinted and cached.
capability
per key
Usage accounting that is not a bill
GET /v1/usage returns synthesized traffic per API key. Why usage data here is operational accounting rather than the source of your monthly invoice.
use case
500
TTS for debt-collection voice agents
Collections dialers speak required disclosures on every connected call. The cost arithmetic of scripted outbound, per-key audit trails, and provenance watermarks.
Your script, this API, thirty minutes — and the recording leaves with you.
Hear it on your own script