Voice Clone Studio — DramaBox Edition

This is a stripped-down version of Voice Clone Studio, rebuilt around DramaBox as the sole TTS engine. Meant for Voice Cloning as well as Lora generation for ComfyUI

For the full-featured version with multiple TTS engines (Qwen3-TTS, VibeVoice, LuxTTS, and more), see the main repo: https://github.com/FranckyB/Voice-Clone-Studio

---

⚠️ DramaBox is highly prone to hallucination. It will frequently add, drop, or mangle words — especially on longer texts. Expect to generate multiple times before getting a clean output. Using a seed lets you reproduce good results once you find them. It is also limited to around 45 seconds

---

A modular Gradio app for local voice workflows powered by DramaBox:

• DramaBox for TTS voice cloning and finetuning
• MMAudio for sound effects
• Qwen3-ASR and Whisper for transcription
• llama.cpp or Ollama for prompt generation

Features

Voice Clone (DramaBox)

• Clone voices from your own samples
• Seeded generation for reproducible outputs
• LoRA support for finetuned speaker voices
• Paragraph splitting for long-form generation
• Prompt Hub integration
• Output metadata tracking

Prep Audio

• Trim, normalize, mono conversion, denoise
• Extract audio from video
• Batch transcription with Qwen3-ASR or Whisper
• Save processed samples and manage datasets

Train Model (DramaBox)

• Finetune DramaBox models from dataset folders
• Progress and log streaming in UI
• Configurable training options

Voice Design

Create voices from natural language descriptions — no audio needed, using Qwen3-TTS Voice Design Model:

• Describe age, gender, emotion, accent, speaking style
• Generate unique voices matching your description

Sound Effects

• Text-to-audio and video-to-audio generation with MMAudio
• Model and generation controls for SFX workflows

Prompt Manager

• Save and reuse prompts
• Generate prompts locally with llama.cpp or Ollama

Output History

• Browse, preview, and manage generated files

Settings

• Enable or disable visible tools
• Configure output and folder paths
• Toggle model engine availability
• Download DramaBox and ASR models

Installation

Prerequisites

• Python 3.11
• Windows or Linux: CUDA GPU recommended for best performance
• macOS: Apple Silicon works with MPS (training can be restricted)
• SOX
• FFmpeg
• Optional: llama.cpp or Ollama

Quick Setup

Windows

setup-windows.bat

Linux

chmod +x setup-linux.sh
./setup-linux.sh

macOS

chmod +x setup-mac.sh
./setup-mac.sh

Note: Manual installation is not recommended. The setup scripts handle vendor dependencies, custom wheels, and module patching that cannot be replicated with a plain pip install.

Usage

python voice_clone_studio.py

Or use launcher scripts:

• Windows: launch.bat
• Linux or macOS: ./launch.sh

Default UI URL: http://127.0.0.1:7860

Project Structure

Voice-Clone-Studio/
+- voice_clone_studio.py
+- config.json
+- requirements_winodws.txt
+- requirements_linux.txt
+- launch.bat / launch.sh
+- setup-windows.bat / setup-linux.sh / setup-mac.sh
+- docs/
+- modules/
   +- core_components/
      +- tools/
         +- voice_clone.py
         +- prep_audio.py
         +- train_model.py
         +- sound_effects.py
         +- prompt_generator.py
         +- output_history.py
         +- settings.py
      +- ai_models/
      +- help_page.py
   +- mmaudio/
   +- qwen_finetune/
   +- vibevoice_asr/

Notes

• The only active TTS engine is DramaBox. The multi-engine support from the main Voice Clone Studio repo (Qwen3-TTS, VibeVoice, LuxTTS) is not included here.
• DramaBox hallucinates frequently. Short, simple sentences work best. For longer texts, use "Split Audio by Paragraph" and review each segment.
• For the full-featured version: https://github.com/FranckyB/Voice-Clone-Studio

License

This project is licensed under Apache 2.0. See LICENSE.

Third-party projects used include:

• DramaBox
• Qwen3-ASR
• Whisper
• MMAudio
• Gradio
• llama.cpp