Voice Studio

Voice Studio is a local desktop app for turning a short reference voice clip into narrated audio on your own machine.

It wraps the open-source VoxCPM voice cloning stack inside an Electron desktop interface so non-technical users can:

• pick a reference voice sample
• paste a post, note, or script
• optionally rewrite the text into a slower spoken narration with Gemini
• generate and review cloned audio locally
• browse previously generated outputs from a built-in history view

Problem This Project Tries to Solve

Voice cloning = generating new speech that sounds like a reference speaker
Voice cloning ဆိုတာ reference voice sample တစ်ခုကိုယူပြီး အဲဒီအသံနဲ့တူတဲ့ speech အသစ်ကို generate လုပ်တာပါ။

Most open-source voice cloning projects are still hard for normal users to run because they usually require:

• command line setup
• Python environment management
• model download steps
• manual script editing
• no desktop workflow for repeated use

This project tries to solve that gap by making local voice cloning feel more like a usable desktop tool instead of a research repo.

What Voice Studio Includes

• Electron desktop app = JavaScript-based desktop shell
  Electron desktop app က web technology ကိုသုံးပြီး desktop app အဖြစ် run လို့ရတဲ့ shell ပါ။
• FastAPI backend = lightweight Python API server
  FastAPI backend က Python နဲ့ရေးထားတဲ့ local API server ပါ။
• VoxCPM runtime = speech generation engine
  VoxCPM runtime က actual voice generation လုပ်ပေးတဲ့ model/runtime ပါ။
• Gemini rewrite preview = optional AI script rewriting before synthesis
  Gemini rewrite preview က text ကို narration style ဖြစ်အောင် optional rewrite လုပ်ပေးတဲ့ feature ပါ။
• History browser = built-in list of previous generated audio files
  History browser က generate လုပ်ပြီးသား audio files တွေကို app ထဲကနေပြန်ကြည့်လို့ရတဲ့ feature ပါ။

Tech Stack

• Electron Desktop window, IPC bridge, file picker, and app packaging
• HTML / CSS / Vanilla JavaScript Renderer UI for the Studio, History, and Setup tabs
• Python Local backend runtime
• FastAPI Local HTTP API between Electron and the voice engine
• VoxCPM Open-source multilingual TTS and voice cloning model stack
• NumPy + SoundFile Audio array handling and .wav export
• Google Gemini API optional AI-assisted narration rewrite and title generation
• uv Python dependency management
• npm / electron-builder Desktop dependency management and macOS packaging

Sources

This project is built on top of these upstream sources:

• OpenBMB/VoxCPM https://github.com/OpenBMB/VoxCPM
• VoxCPM2 model https://huggingface.co/openbmb/VoxCPM2
• VoxCPM documentation https://voxcpm.readthedocs.io/en/latest/
• Google Gemini API https://ai.google.dev/

Project Structure

.
├── backend/             # Local FastAPI server
├── electron/            # Desktop shell and renderer UI
├── VoxCPM/              # Upstream voice generation codebase
├── package.json         # Electron app scripts
└── README.md

Local Setup

1. Install Node.js dependencies

npm install

2. Install Python dependencies for VoxCPM

cd VoxCPM
uv sync
cd ..

3. Prepare model weights

Model weights = trained AI model files used during inference
Model weights ဆိုတာ AI model run ဖို့လိုတဲ့ trained data files တွေပါ။

You have two options:

• use local weights under VoxCPM/pretrained_models/VoxCPM2
• or let the runtime load from openbmb/VoxCPM2 if your environment is configured for that

Important:

• large model files should not be committed to GitHub
• this repo ignores local downloaded weights by default

Run The App

npm start

The Electron app will automatically start the local Python backend.

Build macOS App

Packaging = turning source code into a shareable app build
Packaging ဆိုတာ source code ကို share လို့ရတဲ့ app build အဖြစ်ပြောင်းတာပါ။

npm run pack:mac

For a zipped output:

npm run dist:mac

Usage Flow

1. Open Setup
2. Choose a reference audio file
3. Optionally add the exact transcript of that reference clip for better similarity
4. Paste your target text in Studio
5. Optionally enable Gemini rewrite to convert written text into a more spoken narration style
6. Click Generate
7. Review the output and open the output folder if needed

Open Source Notes

• local model files are ignored from Git
• generated audio outputs are ignored from Git
• local virtual environments and caches are ignored from Git
• Gemini API usage is optional and requires your own API key

Responsible Use

Consent = permission from the original speaker
Consent ဆိုတာ original speaker ဆီက သဘောတူခွင့်ပြုချက် ရထားတာပါ။

Please use this project responsibly:

• only clone voices when you have permission
• do not impersonate people without consent
• clearly disclose synthetic audio when appropriate
• review your local laws and platform policies before publishing generated content

License

This repository is released under the Apache 2.0 License.

Please also review the upstream licenses and notices from VoxCPM and any other third-party dependencies you redistribute.