FulreachHelp
Messages

Voice notes, TTS, and voice cloning

Generating AI voice audio, account voices, and invite-based voice cloning flows.

Last updated: May 2026

Overview

Fulreach generates LinkedIn voice-note compatible audio by orchestrating ElevenLabs text-to-speech, transcoding (commonly MP3 → M4A) for Unipile compatibility, then delivering through messaging send paths.

Account voices

Each workspace seat lists voices linked to the active account (database mapping voice_accounts → voices). Generation routes ensure a voice row exists or upsert minimal metadata before synthesis.

Voice cloning / invites

Creating a custom voice may follow an invite / token lifecycle: owner creates invite, external contributor records samples, Fulreach finalises mapping to Instance Voice Clone (IVC) resources in ElevenLabs—exact UX steps depend on release; treat invite links as single-use & expiring.

Operational expectations

  • Provider latency & quota impact time-to-first-byte for preview vs final send.
  • Failures often surface as 400 validation (bad voice id) or upstream 5xx during generation—retry after cooldown.
  • Always confirm legal consent for voice likeness cloning in your jurisdiction & prospect communication policies.

Automated sequences that send audio steps pull step metadata (text body, voice id, tuning parameters: stability, similarity, speed, style) before encoding—see Sequences and automated follow-ups.