Commit History
-
Adicionar script de avatar em tempo real e benchmark RTF
marcos +1: Claude Opus 4.5- realtime_avatar.py: Avatar com modelos pre-carregados - benchmark_rtf.py executado no servidor mostra: * StyleTTS2 com diffusion_steps=5: RTF=0.04 (22.9x tempo real) * StyleTTS2 com diffusion_steps=3: RTF=0.06 (16.5x tempo real) Resultados do benchmark (RTX 3090): - Audio de 6s gerado em 0.27s - Muito mais rapido que tempo real 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>d805b79 -
Setup completo para MuseTalk V1.5 + StyleTTS2
marcos +1: Claude Opus 4.5Inclui: - install.sh: Script de instalacao automatica - requirements.txt: Versoes testadas e funcionais - scripts/generate_audio.py: Gerar audio com clonagem de voz - scripts/run_lipsync.py: Executar lip sync - scripts/extract_voice_ref.py: Extrair audio de referencia - scripts/full_pipeline.py: Pipeline completo Versoes criticas para compatibilidade: - accelerate==0.25.0 - diffusers==0.21.0 - huggingface-hub==0.25.0 - styletts2==0.1.6 Testado em: vast.ai com CUDA 12.4, Python 3.12, PyTorch 2.6 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>bb3d782