VERSION.md

CosyVoice Version Information

Current Version: v1.0-cosyvoice-300m

Models Installed:

• CosyVoice-300M (Main model)
• CosyVoice-300M-SFT (Supervised Fine-Tuning)
• CosyVoice-300M-direct (Zero-shot inference)
• CosyVoice-ttsfrd (Required resources)

Features:

• Multi-language TTS (Chinese, English, Japanese, Korean)
• Zero-shot voice cloning
• Cross-lingual synthesis
• GPU acceleration with RTX A5000

Performance:

• Generation speed: ~1x real-time
• Model loading: 5-10 seconds
• GPU: RTX A5000 (24GB VRAM)

Known Issues:

• Chinese accent in English/Portuguese synthesis
• Model trained primarily on Chinese data

Next Version:

• CosyVoice2-0.5B (downloading)
• Improved English pronunciation
• Lower latency (150ms)
• 30-50% reduction in pronunciation errors