OpenAI Text to Speech
OpenAI Text to Speech reads out the text you give it with an OpenAI voice and returns an audio file: MP3 by default, WAV or OGG if you choose. Every voice speaks every language: the text sets the language. With the default model you can describe in words how you want it to sound. If the text is long, it splits it at the end of sentences, reads each part and joins them into a single file.
You need an OpenAI API credential with your OpenAI account key. See Credentials. The text is sent to OpenAI to produce the audio.
This is not the voice node for calls: to speak to a caller during a live call, the Voice Agent uses the OpenAI TTS node. This component returns a finished file.
When to use it
- Send your customers a voice note confirming an appointment or an order.
- Produce voice-overs in a particular tone, for example warm and unhurried.
- Let an agent answer with audio, using the component as a tool.
- Produce the audio for a list of texts in one go with Batch Mode.
Configuration
| Field | What it does |
|---|---|
| API Key (required) | Your OpenAI key. The key is filled in for you when you choose your OpenAI credential. |
| Text to Speech (required) | The text to read out. You can type it or connect it from another component. |
| Voice | The voice. Default: coral. The others are alloy, ash, ballad, cedar, echo, fable, marin, nova, onyx, sage, shimmer and verse. With the tts-1 and tts-1-hd models, ballad, cedar, marin and verse are not available. |
| Model | The voice model. Default: gpt-4o-mini-tts, the only one that follows Voice style. tts-1 is faster and tts-1-hd sounds cleaner. |
Advanced options
| Field | What it does |
|---|---|
| Voice style | Optional. How you want it to sound, in words: “warm and unhurried”, “brisk and professional”. Only the gpt-4o-mini-tts model uses it. |
| Speed | The reading speed, from 0.25 to 4. Default: 1, the voice’s natural pace. |
| Audio format | The file format: mp3 (default), wav or ogg. MP3 is small and plays everywhere (WhatsApp, email, browser); WAV is uncompressed; OGG suits voice notes. |
Output
The Data output returns an audio file with the text read out, in the chosen format. Its name is the first words of the text, without accents or punctuation (for example, Hola_buenos_dias_Su.mp3).
Connect it to a component that stores or sends files. An agent can also use the component as a tool to create audio when it needs to; see what tools are in Key concepts.
Example
A gym chain sends its members an encouraging message every Monday. An agent writes the text, OpenAI Text to Speech reads it with the nova voice and Voice style set to “energetic and friendly”, and the audio is sent over WhatsApp with Batch Mode, one per member.
Related components
- Speech to text and text to speech: every component that transcribes audio or produces it.
- OpenAI Whisper: the opposite, turns audio into text with OpenAI.
- Gemini Text to Speech: another way to describe the tone in words.
Tips
- In Voice style, describe only how it sounds, not the message: the text to read always goes in Text to Speech.
- Measured in our test environment on 3 November 2026: the MP3 of the sentence “Hola, buenos días. Su cita está confirmada para el martes a las diez y media.” was produced in 1.5 s.