Voice Foundry

Create a voice library, select the active voice, and use it in Realtime.

OpenAI custom voices require a consent recording first, then the voice object can be reused in TTS and Realtime. This page keeps the voice id, model, and launch notes in one place so the active selection actually flows into the bot.

Library size 0
Custom voices 0
Built-ins 0
Active voice -

Create voice entry

Save a built-in realtime voice or record a new custom voice from audio. Existing voice_... ids still work, and if OpenAI custom voice access is unavailable we keep the draft locally so you can listen back and keep working.

Local fallback is built in If your OpenAI organization does not have custom voice access yet, we still save the recordings, notes, and metadata as a local draft so you can listen back and keep working.
Custom voices use a voice id like voice_123abc. Built-ins match the current realtime flow.
This is the model the realtime session will use when this voice is active.
Used for text-to-speech snippets and any future non-realtime playback.
Paste an existing consent id here, or record a new consent clip below.
Use this as the script for your voice sample and for the TTS preview.
Read the approved consent phrase aloud. We'll upload it if you create a new custom voice.
00:00 Mic idle
No consent recording yet.
Record the voice you want the model to learn, then listen back before saving.
00:00 Mic idle
No sample recording yet.
If checked, the new voice becomes the one the realtime bot uses on the next session.

Voice library

Search, select, and remove voices from the library. The active entry is highlighted and becomes the default for new realtime sessions.

Loading voice entries...