Media & Motion v1.0.0 MIT Verified by NVIDIA SkillSpector

Lumoz Voice Cloning & Multilingual TTS v4

Clone any human voice from a 10–120s audio sample and synthesize expressive speech in 90+ languages (including Bengali) via MCP

Created by Lumoz AI @LumozAI Source code
Install skill
$ npx lumoz-skills add lumoz-voice-cloning

Installs into your repository's .agents/skills/ directory.

$ curl -s https://lumozai.com/api/skills/lumoz-voice-cloning/raw > SKILL.md

Pipes raw SKILL.md into your current directory.

Create .agents/skills/lumoz-voice-cloning/SKILL.md and paste the specification below.

Download raw SKILL.md ↓
5 Stars
42 Installs
4 Agents

SKILL.md specification

---
name: lumoz-voice-cloning
description: Clone any user voice from a 10–120 second audio recording and synthesize ultra-realistic speech in 90+ languages (including Bengali, Hindi, Arabic, English, Spanish, etc.) using Lumoz TTS v4 via MCP.
---

# Lumoz Instant Voice Cloning & Multilingual TTS v4

Yes! **Lumoz AI natively supports Instant Voice Cloning (`lumoz_clone_voice`) and 90+ language Text-to-Speech (`lumoz_generate_tts`)—including Bengali (`বাংলা`)—directly through MCP.**

## 1. How to Clone a User's Voice in MCP

If the user uploads or provides an audio recording of their voice (10 to 120 seconds of clean speech in MP3, WAV, or M4A):

### Option A: Permanent Reusable Voice Clone (`lumoz_clone_voice`)
1. If the user's audio is a local file or chat attachment, first upload it to Lumoz R2 CDN using `lumoz_upload_media` (pass the base64 data URI).
2. Call `lumoz_clone_voice`:
   ```json
   {
     "name": "User Custom Voice",
     "audio_url": "https://cdn.lumozai.com/uploads/...",
     "remove_background_noise": true,
     "sample_text": "লুমোজ এআই ভয়েস ক্লোনিং এখন বাংলায় কথা বলতে পারে।"
   }
   ```
3. `lumoz_clone_voice` returns a permanent white-label `voice_id` (e.g. `"lumoz-voice-user-custom-a8f2k"`).
4. Pass that `voice_id` into `lumoz_generate_tts` for all future voiceovers!

### Option B: 1-Step Instant Clone & Speak (`lumoz_generate_tts` with `reference_audio`)
You can also pass `reference_audio` directly inside `lumoz_generate_tts` (or `lumoz_run_model` with `model_id: "lumoz-tts"`):
```json
{
  "text": "[warmly] নমস্কার, এটি আমার নিজের ক্লোন করা কণ্ঠে তৈরি করা হয়েছে।",
  "reference_audio": "https://cdn.lumozai.com/uploads/..."
}
```

## 2. Directing Emotion & Merging with Video
- **Inline Emotion Tags**: Use bracketed cues in `text` such as `[whispers]`, `[excited]`, `[calmly]`, `[laughs]`, or `[narrator tone]`.
- **Merging Voiceover into Video**: Once `lumoz_generate_tts` returns the `audio_url`, call `lumoz_run_microservice` with `service: "video-audio-merger"`, `video_url`, and `audio_url` to combine the cloned voiceover with any video!