Skip to main content
OpenAI Text-to-Speech converts text into spoken audio using OpenAI’s audio model line. Its defining feature is steerability: alongside the text you supply an instruction describing how it should be delivered, and the model adjusts accordingly. This was the first mainstream TTS API to expose delivery as an instruction rather than as a per-voice parameter set.
Available in Ollang Workflows as openaiTts — AI Dubbing. See the Text-to-Speech catalog.

​
Models and voices

The 13 built-in voices are alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer, verse, marin, and cedar. tts-1 and tts-1-hd support the subset alloy, ash, coral, echo, fable, onyx, nova, sage, and shimmer. Output formats: MP3 (default), Opus, AAC, FLAC, WAV, and PCM.

​
Steering

On gpt-4o-mini-tts, the instructions parameter controls:
  • accent
  • emotional range
  • intonation
  • impressions
  • speed of speech
  • tone
  • whispering
An instruction such as “speak in a warm, reassuring tone with occasional pauses for emphasis” changes the performance without changing the voice. For dubbing this is the practical substitute for line-by-line direction.

​
Where it fits in a workflow

This is the engine for volume with direction: internal communications, training modules, product walkthroughs, compliance content, and first-pass drafts, where delivery needs steering but a bespoke voice identity does not. A useful pattern: run the full backlog through OpenAI TTS, review, and re-run only the segments that need a distinctive voice through Cartesia or ElevenLabs.

​
Trade-offs

  • No custom or cloned voices. A character cannot be carried across a series with a distinctive voice identity the way ElevenLabs, Cartesia, or Azure custom neural voices allow. If brand voice consistency is a requirement, this is the disqualifying constraint.
  • Language support broadly follows Whisper’s 50+ languages, but the voices are optimized for English. Quality outside English is less consistent than the language count suggests — benchmark your specific target languages rather than assuming parity.
  • Narrower expressive ceiling than Gemini TTS or Cartesia on performed content.

​
Reference