Source: OpenAI Text-to-Speech Guide

OpenAI API guide to the Speech endpoint: deprecated TTS models and Realtime migration, built-in voices, instructions steering, output formats and the AI-voice disclosure rule.

Contents
  • What: official API documentation (Text to speech guide)
  • Author/publisher: OpenAI
  • Date: living document; read 2026-11-14
  • Original: Text to speech

Summary

Describes /v1/audio/speech, its voices and formats, and the move of TTS onto the Realtime API.

Key claims

  • tts-1, tts-1-hd and gpt-4o-mini-tts shut down 2027-01-06; the migration target is gpt-realtime-2.1-mini via the Realtime API, which is a workflow change rather than a model swap.
  • tts-1 trades quality for lower latency than tts-1-hd.
  • instructions on gpt-4o-mini-tts steer accent, emotion, intonation, speed, tone, whispering.
  • 13 built-in voices (marin and cedar recommended); voices are tuned for English but language coverage follows Whisper.
  • Output: mp3 (default), opus, aac, flac, wav, pcm (raw 24 kHz 16-bit LE). WAV/PCM give the fastest response.
  • Audio streams via chunked transfer encoding so playback starts before generation ends.
  • Custom voices require a consent recording from the speaker.
  • Usage policy requires disclosing to end users that the voice is AI-generated.

Feeds

Voice agents · Multimodal models

Suggest changes →