Text to Speech

POST https://api.mafdet.ai/v1/audio/speech

Turn text into natural speech with the Gemini TTS models. The endpoint is OpenAI-compatible, so any OpenAI SDK works by pointing base_url at https://api.mafdet.ai/v1. The response body is the audio file itself (binary), not JSON.

Both are Gemini 2.5 preview TTS models served through Vertex AI.

Request fields

These are Gemini voices — OpenAI voice names (alloy, nova, …) are rejected with TTS_VOICE_NOT_ALLOWED.

Response

The raw audio bytes: 24 kHz, 16-bit, mono. With wav you get a standard RIFF/WAVE file; pcm returns the bare samples.

Billing

TTS is billed on tokens, both directions: the text input tokens plus the audio output tokens, at the per-model rates above. A short sentence typically produces 40–80 audio output tokens (well under a tenth of a cent). Audio output is capped at 4,000 tokens per request. A rejected request (bad voice, text too long, …) returns 400 before the provider is called and is never billed; a failed provider call releases its reservation in full.

curl

curl https://api.mafdet.ai/v1/audio/speech \
  -H "Authorization: Bearer $MAFDET_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-2.5-flash-tts",
    "input": "Hello from Mafdet AI.",
    "voice": "Kore",
    "response_format": "wav"
  }' \
  --output speech.wav

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.mafdet.ai/v1",
    api_key="sk-mafdet-xxxxxxxxxxxxxxxx",
)
audio = client.audio.speech.create(
    model="gemini-2.5-flash-tts",
    input="Hello from Mafdet AI.",
    voice="Kore",
    response_format="wav",
)
audio.write_to_file("speech.wav")

Node.js

const res = await fetch("https://api.mafdet.ai/v1/audio/speech", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.MAFDET_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "gemini-2.5-flash-tts",
    input: "Hello from Mafdet AI.",
    voice: "Kore",
    response_format: "wav",
  }),
});
const buf = Buffer.from(await res.arrayBuffer());
await import("node:fs/promises").then((fs) => fs.writeFile("speech.wav", buf));

Error codes

All of the above return 400 before the provider is called, so a rejected request is never billed.

Next steps

• Models overview
• API basics