Why it is worth attention

It offers a complete, guided pipeline from audio extraction to voice generation entirely on your own GPU, with strong privacy guarantees and a zero-shot mode for instant cloning without training.

Who it is for

Developers building custom voice interfaces
Content creators needing personalized narration
Researchers exploring few-shot voice adaptation
Privacy-conscious users avoiding cloud services

Use cases

Cloning your own voice for a digital assistant
Generating audiobooks in a personalized style
Voice dubbing for personal video projects
Creating voiceovers for accessibility tools

Strengths

Fully local operation ensures data never leaves your machine
One-command pipeline handles extraction, cleaning, transcription, and training
Zero-shot mode enables instant voice cloning from a single clip
Includes a web UI with four tabs mirroring all CLI capabilities

Considerations

Requires Linux with an NVIDIA CUDA GPU and at least 24 GB VRAM for training
Voice quality heavily depends on clean source audio and careful transcript proofreading
Currently relies on Qwen3-TTS base model and may not work on non-CUDA hardware

README quick start

Voice Clone Lab

Clone a voice from a few minutes of audio and generate speech from text — locally, on your own GPU. Give it 5–15 minutes of clean speech (a recording, or a YouTube link) and it fine-tunes Qwen3-TTS into a personal voice model you can drive from a CLI or a web UI.

Any source: upload a recording, record from your mic, or paste a YouTube/URL link
Guided pipeline: extract → clean → split → transcribe → proofread in an editable table → train
Your voices, organized by name: train as many as you like, pick one and type text
Zero-shot mode: try a voice instantly from a single clip, no training
Fully local: audio, transcripts, and voice models never leave your machine

Consent notice

Use this project only for your own voice or a voice where you have explicit written permission from the speaker. Do not use it for impersonation, fraud, bypassing consent, harassment, or cloning public figures or private people without permission. The same rule applies to URLs: only download content you have the rights to use.

Requirements

Linux with an NVIDIA CUDA GPU
Training: 24 GB VRAM minimum (RTX 3090/4090 class, batch-size 2). 32 GB+ is comfortable.
Generation only (using a checkpoint someone else trained): any ~8 GB card works.
Python ≥ 3.10 (3.12 recommended), ffmpeg, git

Install

git clone https://github.com/tetsuo-ai/voice_clone_lab.git
cd voice_clone_lab

pip install -e '.[ui]'     # core + web UI; drop [ui] for CLI-only
vcl setup --download       # fetch Qwen3-TTS (pinned + patched) and the 1.7B base weights (~4 GB)
vcl check                  # verify your machine is ready

Optional extras: pip install -e '.[asr,yt,vad,denoise,dev]' — WhisperX (word timestamps), yt-dlp (URL sources), WebRTC VAD (better chunking), mild denoising, and the test suite.

Quickstart

1. Get some audio of your voice

Record 5–15 minutes in a quiet room, one mic, natural reading pace (ready-made reading text: recording_scripts/voice_dataset_reading_script.txt). Or skip recording and paste a link to something you're in:

vcl run --speaker alex --input "https://www.youtube.com/watch?v=…"
# or a local file (wav/mp4/m4a/mp3):
vcl run --speaker alex --input ~/recordings/alex.wav

That one command extracts, cleans, splits, transcribes, and

Description

Clone a voice from a few minutes of audio and generate speech locally — Qwen3-TTS fine-tuning pipeline with CLI and web UI

Related repositories

Similar projects matched by category, topics, and programming language.

lopopolo

Featured

harness-engineering

Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.

AI & Machine LearningAI Agents

2,390

slvDev

Featured

esp32-ai

A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.

AI & Machine LearningLarge Language Models

1,960

littledivy

Featured

mimic

mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.

AI & Machine Learning

1,482