{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreih6aq2oomqwaxxbqd437gjlu4qucudttbwa7en2rjyb3xrbvdcjr4",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mlmtjko7mqi2"
  },
  "path": "/t/tts-with-a-question-the-audio-back-has-the-answer/1377233#post_9",
  "publishedAt": "2026-05-12T03:04:45.000Z",
  "site": "https://community.openai.com",
  "tags": [
    "documentation"
  ],
  "textContent": "OpenAI_Support:\n\n> gpt-audio-1.5 is also worth testing\n\nYou will note that documentation clearly indicates this is a Chat Completions-only multimodal AI model:\n\nYou will have to prompt Chat Completions audio models extremely well yourself with recitation as the task that the AI is doing, and containerize and delineate the text to be spoken strong enough to avoid instruction injection, because gpt-audio-1.5 is _specifically for_ chatting with someone and following user instructions and performing user language tasks.\n\n…just as the internal prompting by OpenAI to make `gpt-4o-mini-tts` into a “speak the text” model is not working well and exhibits this forum topic’s symptom.",
  "title": "TTS with a question... the audio back has the answer"
}