# Voice Clone Lab

A Python tool that fine-tunes Qwen3-TTS on a few minutes of audio to clone a voice and generate speech locally on your own GPU, from a CLI or web UI.

Voice Clone Lab fine-tunes Qwen3-TTS into a personal voice model from 5 to 15 minutes of clean speech, sourced from a recording, a microphone capture, or a YouTube or URL link. Once trained, the voice can be driven from a command-line tool or a web UI, and a zero-shot mode can try a voice instantly from a single clip without training.

The pipeline extracts, cleans, splits, transcribes and lets the user proofread a transcript in an editable table before training, and multiple trained voices are organized by name. All audio, transcripts and voice models stay on the local machine.

It requires a Linux machine with an NVIDIA CUDA GPU: at least 24 GB VRAM for training, or about 8 GB for generation only from an existing checkpoint. The project restricts use to a person's own voice or a voice with explicit written permission from the speaker, and prohibits impersonation, fraud or cloning public or private individuals without permission. It is written in Python and licensed under Apache-2.0.

Website: https://github.com/tetsuo-ai/voice_clone_lab