Model Selection Guide

Voconly supports multiple ASR models. Choose based on your use case and hardware.

Recommended Models for Beginners

SenseVoice Small

Best for: Pure Chinese scenarios

• Excellent Chinese recognition
• Fast response
• Lightweight and efficient

Qwen-ASR 1.7B

Best for: Chinese-English mixed scenarios

• Higher recognition accuracy
• Better for mixed language content
• Requires more VRAM and computing resources

The system downloads the Q5 quantized version by default as a balance between performance and quality.

Downloading Additional Models

If you have a local GPU and ample memory, you can download more quantized versions:

• ModelScope (Chinese users): https://www.modelscope.cn/profile/voconly
• HuggingFace (English users): https://huggingface.co/voconly-org

Quantization Quality Ranking

F16 > Q8 > Q5 > Q4

Earlier versions are larger and use more resources—choose based on your hardware.

Importing External Models

1. Download the model files
2. Place them in the default location or a custom location
3. In Settings → Model List → Import External Model
4. Add the model directory

Note: Each model uses the highest quantization version by default—if both Q5 and Q8 exist, Q8 will be used.

Next Steps

• Shortcut Settings
• Scenario Configuration