How to clone a voice offline on Windows

Install Mimic, download the Qwen3-TTS 0.6B engine from inside the app, record yourself reading a short passage, then type what the voice should say and press Ctrl Enter. It needs about 2.5 GB of VRAM, or about 6 GB of RAM with no GPU, and nothing is uploaded anywhere.

What you need

Windows 10 or 11. Mimic also runs on Linux; there's no Mac version yet.
An NVIDIA graphics card helps, but isn't required: Qwen3-TTS also runs on the CPU. See voice cloning without a GPU.
Two free command-line tools: uv, which downloads the engines, and ffmpeg, which imports and converts audio.
A clean recording of the voice: 3 to 30 seconds of one person talking, or you reading the passage Mimic shows you.

Step by step

Install Mimic. Download Mimic_<version>_x64-setup.exe from the latest release and run it.

Install uv and ffmpeg. In a terminal:

winget install astral-sh.uv
winget install Gyan.FFmpeg

Download an engine. Open Settings › Engines and download Qwen3-TTS 0.6B. Download speech-to-text from the same list too, so transcripts write themselves.
Add a voice. Press Record, type your name and read the passage aloud (8 to 20 seconds), or drop a 3 to 30 second clip onto the window. Mimic transcribes it on your PC.
Generate. Type a script in the middle and press Ctrl Enter. Press Space to listen.
Keep it. Press Ctrl S to save the take as WAV, FLAC or MP3. Unsaved takes are cleared when Mimic closes.

Mimic with the Narrator voice selected, a script in the middle and two takes on the right, Qwen 0.6B loaded using 5.1 of 15.9 GB of VRAM — Voices on the left, the script in the middle, takes on the right. The bar at the top shows real VRAM use.

Getting a good clone

Use one speaker, close to the microphone, with no music or background noise.
Generate the same line a few times: each take is a little different. Compare engines makes it with every engine that fits, and B hides which is which.
Qwen3-TTS needs the exact words of the clip. Recording the passage, or downloading speech-to-text, takes care of that.

Only clone voices you have permission to use. Your own is fine; for anyone else, ask first. Mimic asks you to confirm this before your first clone and marks saved takes as AI-generated. Using a cloned voice to impersonate, deceive or defraud someone is illegal in many places.

Updated 23 October 2026 · Measured with Mimic on an RTX 5070 Ti and a 16-thread CPU, unless marked as published.