Custom voices
Your voice, on every call.
Clone a voice from one clean sample. The agent then speaks in your own on-brand voice, with that voice fixed for the full call.
0.56s median first response · internal benchmark, July 2026
How it works
A consented voice, enrolled once and governed throughout.
Enroll the speaker from a single clean recording. If the source contains two people, the system separates the two speakers and clones only the enrolled one.
source recording → two speakers detected
speaker separation → enrolled speaker isolated
other speaker → excluded from enrollment
clean voice sample ready
A voice is pinned per campaign before calls begin. The call keeps that voice from opening disclosure through hang-up; it never changes voice mid-conversation.
campaign → claimant intake
voice → enrollment approved
call start → voice lock applied
mid-call voice changes blocked
Speech-to-text, reasoning, and synthesis each run on the strongest available vendor for that part of the call. Provider failover keeps the stack from depending on one engine.
speech-to-text → best available route
reasoning → best available route
synthesis → enrolled campaign voice
vendor failure → failover route ready
Voice enrollment is conditioned on the speaker's documented consent: our terms require it, and you keep that consent on file. The agent discloses that it is AI on every call, regardless of how natural the voice is. Caller audio is recorded and transcribed under your settings; no voiceprints or biometric identifiers are created from callers.
speaker consent → required by the terms of use
AI identity disclosure → every call
caller voiceprints → not created
consent stays on file with your organization
source recording → two speakers detected
speaker separation → enrolled speaker isolated
other speaker → excluded from enrollment
clean voice sample ready
campaign → claimant intake
voice → enrollment approved
call start → voice lock applied
mid-call voice changes blocked
speech-to-text → best available route
reasoning → best available route
synthesis → enrolled campaign voice
vendor failure → failover route ready
speaker consent → required by the terms of use
AI identity disclosure → every call
caller voiceprints → not created
consent stays on file with your organization
Thirty minutes. Hear your own voice.
We'll run a live call, work a sample docket, and show you the system refusing a call it shouldn't make.