How to set the tone of an OpenAI voice with instructions
Pick an OpenAI voice in Text to Speech, tap a preset such as News or Story, or describe the pace and tone in your own words. Instructions don't change the price.
Updated November 2, 20263 min read
An instruction tells the voice how to speak: slower, warmer, like a news anchor or like a fairy-tale narrator. The Instructions block appears on the Voice tab only when an OpenAI voice is selected. MiniMax and ElevenLabs voices have their own settings instead, described at the end of this article.
How to add an instruction
In Text to Speech, enter the text and tap Continue.
On the Voice tab, open the OpenAI tab and choose a voice, for example Coral or Onyx.
Under the list, in the Instructions block, tap a preset. Its text appears in the field below, and you can edit it.
Or type your own description in the field "Describe how the voice should sound...".
Tap Continue, then Generate.
To remove the preset, tap it again. The field is cleared.
What the presets do
There are 8 presets. If the app is in Russian, the preset text is inserted in Russian, in other languages it is in English.
| Preset | What it asks the voice to do |
|---|---|
| Calm | Speak calmly and naturally, at a measured pace with natural pauses. |
| Cheerful | Sound cheerful and enthusiastic, with a smile in the voice, a slightly upbeat pace and lively intonation. |
| News | Speak like a news anchor: clear articulation, neutral intonation, a confident and steady pace. |
| Whisper | Speak in a soft, gentle whisper, as if sharing a secret. |
| Story | Sound like a warm audiobook narrator who creates atmosphere and stresses key moments. |
| Teacher | Speak plainly, pause after key points and stress important words. |
| Dramatic | Sound like a movie trailer voice-over: deep tone, slow pace, long pauses. |
| Kids | Sound friendly, playful and warm, speak slowly with a smile and highlight fun words. |
How to write your own instruction
Describe the pace, tone, emotion and, if needed, the accent or role separately. The more specific the description, the closer the result. A few examples:
"Slow pace, slight British accent, thoughtful pauses between sentences."
"Friendly support agent: warm, patient, smiling."
"Sports commentator: fast, excited, louder at the end of each phrase."
An instruction is a request to the voice, and the result can differ between voices and languages. If the voice ignores it, make it shorter and more concrete, or try another OpenAI voice. Each new generation is charged again.
Price and saving
An instruction doesn't change the price: an OpenAI voice costs 1.8 tokens for every started 150 words of your text, with or without it. The app doesn't keep your instruction for the next session. When you open a result from History, the instruction it was made with is restored. To reuse instructions later, keep them in a notes app.
What to use with other voices
| Voices | Instead of instructions |
|---|---|
| MiniMax, including your own MiniMax voice | Emotion, Speed and Pitch under the list; tags such as (laughs) and a pause <#1#> in the text |
| ElevenLabs | Stability, Similarity and Style; tags such as [whispers] or [excited] in the text |
| Qwen 3, including your own Qwen 3 voice | No settings |
A comparison of all voice groups: Text to Speech voices.
Contact support
Describe the problem and we will reply by email. Add the file name or job ID if you have one.