Use when cloning a voice locally (offline, free, no API) from 3 seconds of reference audio with Qwen3-TTS. Triggers on "clonar voz local", "voice clone offline", "alternativa ElevenLabs grátis", "Qwen3-TTS", "TTS local Apache 2.0".
Scanned 9/19/2026
Install to Claude Code
npx -y skills add majinmagros/magros.ai-skills --skill voice-cloning-local --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Voice Cloning Local?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/majinmagros-voice-cloning-local)More formats (shields.io, HTML) on the badges page.
---
name: voice-cloning-local
description: Use when cloning a voice locally (offline, free, no API) from 3 seconds of reference audio with Qwen3-TTS. Triggers on "clonar voz local", "voice clone offline", "alternativa ElevenLabs grátis", "Qwen3-TTS", "TTS local Apache 2.0".
metadata:
origin: ECC
validated: 2026-08-22
source_video: npVm4tBalp8
official_sources:
- https://huggingface.co/spaces/Qwen/Qwen3-TTS
- https://github.com/Qwen-TTS/Qwen3-TTS
- https://localaimaster.com/blog/qwen3-tts-local-setup (2026-08-16)
---
## Pipeline Qwen3-TTS (Voice Cloning Local)
Qwen3-TTS (Alibaba, Apache 2.0): **voice cloning** (3s de referência → qualquer texto), **voice design** (descrição natural), 10 idiomas, modelos 0.6B (1.8GB) e 1.7B (3.9GB) — roda em GPU consumer, 100% offline.
## Quando usar
- Clonar voz localmente sem API, grátis, offline
- Criar voz via descrição natural (idade, gênero, estilo)
- Batch de narrações com a mesma voz clonada
- Servidor local com Web UI para testes rápidos
## Quando NÃO usar
- Síntese genérica sem referência → use `fal-ai-media`
- Streaming TTS real-time ou 50+ idiomas (Qwen suporta 10)
### Validação Oficial (2026-08-22)
| Claim | Status | Fonte |
|---|---|---|
| Apache 2.0 | ✅ | HF model cards, GitHub LICENSE |
| Clone 3s / roda local (`pip install qwen-tts`) | ✅ | Base models, vLLM-Omni, ComfyUI |
| 10 idiomas | ✅ | Model card |
| Similarity 0.789 vs ElevenLabs 0.75/MiniMax 0.72 | ✅ | Benchmark (`references/benchmark-comparison.md`) |
---
## Passo a Passo
### 1. Ambiente
```bash
pip install -U qwen-tts # simples; + FFmpeg; Python 3.10+, CUDA 4GB+ (0.6B) / 8GB+ (1.7B)
# Pesos manuais:
huggingface-cli download Qwen/Qwen3-TTS-Tokenizer-12Hz --local-dir ./Qwen3-TTS-Tokenizer-12Hz
huggingface-cli download Qwen/Qwen3-TTS-12Hz-1.7B-Base --local-dir ./Qwen3-TTS-12Hz-1.7B-Base
```
### 2. Clone / Design / Batch (scripts prontos em `scripts/`)
```bash
python scripts/voice_clone.py '<texto>' ref.wav '<transcrição EXATA do ref>' Portuguese 1.7B
python scripts/voice_design.py # voz via descrição natural
python scripts/tts_cli.py clone "Texto" ref.wav "Transcrição" --lang Portuguese --model 1.7B
python scripts/tts_cli.py design "Texto" "young female, warm, Portuguese" --lang Portuguese
python scripts/batch_clone.py # múltiplos textos, mesma voz
```
Regras: `ref_text` = match EXATO do áudio (crítico p/ qualidade); ref limpo 10-30s, 16kHz+; cross-lingual funciona mas pode vazar sotaque — prefira ref no idioma alvo.
### 3. Web UI local (`http://localhost:7861`, modelo em cache)
```bashIs this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!