Open-Source Model Infrastructure

Supported AI Models

Miralas.io utilizes category-leading open-source models for ultra-low latency voice agents and zero-shot voice cloning. Data privacy and speed are our top priorities.

The Brain / LLM

Meta Llama 3.1 8B

Dialogue, Reasoning & Logic Engine

A highly optimized large language model fine-tuned for real-time conversational AI, delivering rapid reasoning and natural dialogue.

Parameters

8 Billion (8B) — optimized for extremely high Tokens Per Second (TPS).

Context Window

128,000 tokens — perfect for remembering long conversations.

Speed & Latency

Time To First Token (TTFT) <200ms on optimized hardware (L40S/H100), crucial for real-time voice agents.

Language Support

Native support for 8 languages (English, German, French, Italian, Portuguese, Hindi, Spanish, Thai). Highly capable in tool-use and API calling.

The Voice / TTS

Resemble AI Chatterbox

TTS & Zero-Shot Voice Cloning Engine

A powerful text-to-speech and voice cloning model trained on hundreds of thousands of hours of audio for natural, expressive speech.

Architecture

~0.5B parameters, trained on 500,000 hours of audio data.

Zero-Shot Voice Cloning

Clone any voice with just a 3–5 second clean audio reference.

Emotion & Prosody Control

Context-aware speech synthesis — whispering, excitement, pacing, and more.

Language Support

Multilingual support across 23+ languages with cross-lingual voice cloning (e.g., clone an English speaker and make them speak fluent Japanese).

Comparison Summary

Feature	Meta Llama 3.1 (8B)	Chatterbox
Primary Role	Text / Logic / Dialogue	Audio / TTS / Cloning
Size	8 Billion	~0.5 Billion
Native Languages	8	23+
Input Needed	Text Prompt	Text + 3s Audio Ref
Output Generated	Text Output	High-Fidelity Audio

Quick Start: Voice Cloning API

Get started with Miralas.io's voice cloning API in just a few lines of TypeScript. Provide a short audio reference and generate speech in any supported language.

typescript

import fs from "fs";

async function cloneVoiceAndSpeak() {
  const audioRef = fs.readFileSync("./my-voice-sample.wav");
  const base64Audio = audioRef.toString("base64");

  const response = await fetch("https://api.miralas.io/v1/audio/speech", {
    method: "POST",
    headers: {
      "Authorization": "Bearer YOUR_API_KEY",
      "Content-Type": "application/json"
    },
    body: JSON.stringify({
      model: "chatterbox-multilingual",
      input: "Merhaba, Miralas altyapısı ile sesimi anında klonlayabilirim!",
      voice_reference: `data:audio/wav;base64,${base64Audio}`,
      language: "tr"
    })
  });

  const result = await response.blob();
  // Save or play the generated audio buffer...
}