Views
No views yet
checkpoints_backup/
├── llm_epoch24_loss0.056_acc99.2.pt (1.9GB)
└── flow_epoch40_loss0.519.pt (430MB)
hf_checkpoints/
├── llm/epoch_24_whole.pt
└── flow/epoch_40_whole.pt
scripts/
├── generate_full_text_v2.py (推荐使用)
├── generate_episode_v6_segmented.py
└── generate_fixed_v2.py
config/
└── cosyvoice2_small_batch.yaml
data/
└── episode_script_v6.txt (示例脚本)1from cosyvoice.cli.cosyvoice import CosyVoice2
2import torch
3
4# 加载基础模型
5cosyvoice = CosyVoice2(pretrained/CosyVoice2-0.5B)
6
7# 加载 fine-tuned checkpoints
8llm_ckpt = torch.load(checkpoints_backup/llm_epoch24_loss0.056_acc99.2.pt)
9flow_ckpt = torch.load(checkpoints_backup/flow_epoch40_loss0.519.pt)
10
11cosyvoice.model.llm.load_state_dict(llm_ckpt, strict=False)
12cosyvoice.model.flow.load_state_dict(flow_ckpt, strict=False)1# 完整文本一次性生成(推荐)
2text = "你的播客文本内容..."
3speaker = "dumubuchenlin"
4
5all_audio = []
6for output in cosyvoice.inference_sft(text, speaker, stream=False, speed=1.0):
7 audio = output[tts_speech].numpy().flatten()
8 all_audio.append(audio)
9
10final_audio = np.concatenate(all_audio)