Speech

POST /v1/audio/speech turns text into spoken audio. It follows the OpenAI text-to-speech format, so client.audio.speech.create works with the base URL changed. Speech models are listed on the models page — among them tts-1, tts-1-hd, eleven-multilingual-v2, eleven-v3 and gemini-2.5-flash-preview-tts.

Generate audio

curl https://api.cloakray.com/v1/audio/speech \
  -H "Authorization: Bearer $CLOAKRAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "input": "Your order has shipped and will arrive on Thursday.",
    "voice": "alloy",
    "response_format": "mp3"
  }' \
  --output speech.mp3

import fs from "node:fs";
import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.cloakray.com/v1", apiKey: process.env.CLOAKRAY_API_KEY });

const audio = await client.audio.speech.create({
  model: "tts-1",
  input: "Your order has shipped and will arrive on Thursday.",
  voice: "alloy",
});

fs.writeFileSync("speech.mp3", Buffer.from(await audio.arrayBuffer()));

The response body is the audio file itself, not JSON. Save the bytes.

Parameters

Parameter	What it does
`model`	Required. A speech model.
`input`	Required. The text to speak.
`voice`	The voice. Each model family has its own voices; without one, the model's default is used.
`response_format`	`mp3`, `wav`, `opus`, `aac`, `flac` or `pcm`, where the model supports it. Defaults depend on the model.
`speed`	Playback speed, on models that support it.
`instructions`	A style direction such as "speak warmly and slowly", on models that take one.

Pricing

Speech is priced per character of input (or per token on some models), shown on each model's page. A long text costs proportionally more, so trim markup and boilerplate before sending it.

Not offered

The response is buffered and returned in one piece; streaming audio as it is generated is not offered. For live, two-way voice, realtime sessions are not offered yet.