Windows 10 / 11 · x64

LLMVoice for Windows

An isolated, verified, and repeatable Windows installation for local voice cloning.

Stable release · v0.1.7
Install with PowerShellView GitHub Release
Requirements

Before you begin

Windows 10 / 1164-bit
Python 3.11–3.1464-bit installation
Approximately 4 GBFree storage
NVIDIA GPU recommendedCPU supported, but slower

FFmpeg Shared is required. The CUDA Toolkit is not; NVIDIA CUDA acceleration is selected automatically when supported.

Installation

Three short command groups.

Open PowerShell as your normal user. Run the commands in order, reopening the terminal after FFmpeg installation.

01FFmpeg Shared
winget install --id Gyan.FFmpeg.Shared -e
02LLMVoice installer
irm https://llmvoice.yusufdere.com/install.ps1 -OutFile install.ps1
powershell -NoProfile -ExecutionPolicy Bypass -File .\install.ps1
03First validation
llmvoice --version
llmvoice doctor
Installer behavior

Kept separate from your system.

Verifies Windows x64 and Python 3.11–3.14
Downloads the release manifest and SHA256-verified wheel
Creates a versioned isolated runtime under %LOCALAPPDATA%\LLMVoice\runtime
Selects CUDA or CPU based on supported hardware
Validates FFmpeg shared DLL bootstrap and synthetic audio decoding
Publishes launcher and install state atomically
Adds only %LOCALAPPDATA%\LLMVoice\bin to User PATH
Preserves voices, models, cache, and configuration

The installer does not modify machine PATH, pollute global Python, or install the CUDA Toolkit.

Runtime profiles

Automatic selection for your machine.

NVIDIA CUDA

CUDA compute runtime for torch and torchaudio on supported NVIDIA systems, with CPU TorchCodec for Windows media compatibility.

CPU

CPU runtime when CUDA is unavailable or explicitly selected. Long-form generation is significantly slower.

First run

Open a new terminal.

A new terminal picks up the PATH change. First synthesis may need network access to download XTTS-v2 and present its CPML terms.

llmvoice --version
llmvoice doctor
License note

MIT source. Separately licensed model.

LLMVoice source is MIT licensed. XTTS-v2 model weights are licensed separately under CPML, and commercial model rights do not come from the MIT license.

Licensing guide