测试环境 Test Environment
- GPU: NVIDIA RTX 4090D
- fast-vc Version: 0.1.0
- Test Date: 2025-06-15
测试参数 Test Parameters
- 采样率 Sample Rate: 16000 Hz
- 音频格式 Audio Format: 16-bit PCM
- 模型 Model: seed-vc v1.0 tiny
- Chunk Size: 500ms
- 扩散步数 Diffusion Steps: 10
性能指标 Performance Metrics
| 并发数量 | Worker | 首token延迟 | 端到端延迟 | 平均chunk延迟 | 平均RTF | 中位数RTF | P95 RTF |
|---|---|---|---|---|---|---|---|
| Concurrency | Worker | First Token Latency | End-to-End Latency | Avg Chunk Latency | Avg RTF | Median RTF | P95 RTF |
| 1 | 6 | 136.0 | 143.0 | 105.0 | 0.21 | 0.22 | 0.24 |
| 3 | 6 | 205.3 | 168.0 | 148.4 | 0.32 | 0.28 | 0.55 |
| 6 | 6 | 133.8 | 241.2 | 177.4 | 0.36 | 0.36 | 0.43 |
| 6 | 12 | 151.5 | 232.3 | 206.3 | 0.42 | 0.43 | 0.50 |
| 8 | 12 | 159.8 | 318.0 | 232.9 | 0.48 | 0.48 | 0.61 |
| 10 | 12 | 144.8 | 351.7 | 217.1 | 0.44 | 0.45 | 0.52 |
| 12 | 12 | 140.1 | 256.6 | 216.6 | 0.44 | 0.45 | 0.51 |
时间单位: 毫秒 (ms)
Time Unit: Milliseconds (ms)
✅ 所有测试场景均达到实时处理要求 (RTF < 1.0)
✅ All test scenarios meet real-time processing requirements (RTF < 1.0)
测试结论 Test Conclusions
- RTX 4090D 推荐生产配置: 可稳定支持12并发。
- RTX 4090D Recommended Production Configuration: Can stably support 12 concurrent requests.
ps:
- 你可以运行
python examples/websocket/concurrent_ws_client.py来测试fast-vc在你本地的性能表现。 - You can run
python examples/websocket/concurrent_ws_client.pyto test the performance of fast-vc on your local machine.