Mimic guides

Practical answers about running voice cloning on your own PC, built on figures measured with Mimic. Each guide says what was measured and on which hardware.

• How to clone a voice offline on WindowsClone a voice on your own Windows PC with Mimic and Qwen3-TTS: install, record ten seconds, type a script, and save the take. About 2.5 GB of VRAM, or 6 GB of RAM with no GPU.
• Voice cloning without a GPUQwen3-TTS 0.6B clones a voice on the CPU with 5.6 GB of RAM, taking about 4.7 seconds per second of speech on a 16-thread CPU. Every engine's CPU needs.
• Qwen3-TTS 0.6B vs 1.7BQwen3-TTS 0.6B needs 2.4 GB of VRAM, 1.7B needs 4.7 GB. On an RTX 5070 Ti both make a second of speech in under 2 seconds; on the CPU 0.6B is faster.
• Which local TTS engines allow commercial use?Licenses of nine local voice cloning engines: Qwen3-TTS, Chatterbox Turbo, VoxCPM2, Audio8 and NeuTTS Air allow commercial use; Breeze TTS 2 and Fish S2 Pro don't; KittenTTS 2 has conditions.