PPactDocs
Voice MCP

Voice tiers

Choose the voice Pact speaks in — Standard (Amazon Polly, no key), Pro (Cartesia, BYOK), or Premium (ElevenLabs, BYOK) — with an automatic fail-safe fallback to Polly.

Voice MCP ships three text-to-speech tiers. All three are wired end to end into the live call path — the greeting for a recognized caller, each turn, action confirmations, and the hangup all speak in the selected tier. (The greeting for an unrecognized caller uses Standard Polly.)

The three tiers

TierProviderKey requiredHow it's delivered
StandardAmazon PollyNoneTwilio speaks it natively via <Say>
ProCartesiaBYOKSynthesized audio played via <Play>
PremiumElevenLabsBYOKSynthesized audio played via <Play>
  • Standard is the default and needs no setup. Amazon Polly is served natively by Twilio through <Say>, so there's no separate synthesis step and no key to manage.
  • Pro uses Cartesia. Bring your own Cartesia API key. Pact synthesizes the reply and plays a cached audio blob via <Play>.
  • Premium uses ElevenLabs. Bring your own ElevenLabs API key, delivered the same cached-<Play> way.

Pro and Premium are bring-your-own-key

The Pro and Premium tiers call Cartesia and ElevenLabs directly with your API key. Add the key on the integration tiles under your admin settings before selecting the tier. Without a valid key, the tier can't synthesize.

Voice selection

On Pro and Premium you can pick a specific voice — either a curated preset or a voice id you paste from Cartesia or ElevenLabs.

Selection, not cloning

Voice tiers let you select a provider voice. There is no in-product voice-cloning flow — Pact does not record or reproduce a custom brand voice from a sample. You choose from presets or paste a provider voice id.

Fail-safe fallback

The synthesized tiers never break a call. If the BYOK key is missing, the text is too long to synthesize, or the provider returns any error, Pact silently falls back to Standard (Amazon Polly via <Say>) for that reply.

A call never fails on TTS

Pro and Premium degrade to Polly on any error or missing key. The caller always hears a reply — at worst in the Standard voice — rather than dead air.

What's next