yousheng

Fine-tuned VoxCPM2 checkpoint (full finetune, step 6000).

推理

```bash git clone https://github.com/OpenBMB/VoxCPM.git cd VoxCPM && pip install -e . huggingface-cli download jing96963/yousheng --local-dir ./ckpt python scripts/test_voxcpm_ft_infer.py
--ckpt_dir ./ckpt
--text "你好，这是 VoxCPM 的微调测试。"
--output out.wav ```

或 Python 中：

```python from voxcpm import VoxCPM import soundfile as sf

model = VoxCPM.from_pretrained("jing96963/yousheng", load_denoiser=False, optimize=True) wav = model.generate( text="你好，这是 VoxCPM 2 的推理测试。", cfg_value=2.0, inference_timesteps=10, max_len=600, normalize=False, denoise=False, ) sf.write("out.wav", wav, model.tts_model.sample_rate) # sr=48000 ```

注意

• dtype 在 config.json 里设为 bfloat16，与训练精度一致
• 输出采样率 48000 Hz
• 不包含 optimizer.pth / scheduler.pth / training_state.json（推理用不到）