fishaudio / fish-diffusion

An easy to understand TTS / SVS / SVC framework
https://diff.fish.audio
MIT License
662 stars 87 forks source link

colab 微调训练那裏出錯 #30

Closed upright2003 closed 1 year ago

upright2003 commented 1 year ago

87 预训练模式有跑通,但是下一步 微调训练那裏出錯 IsADirectoryError: [Errno 21] Is a directory: '/content/fish-diffusion/checkpoints'

88

請問該如何解決? 謝謝

Mixomo commented 1 year ago

Same error here!

leng-yue commented 1 year ago

你需要手动制定预训练的文件名 (可以在 release 下载), 而不是目录.
You should specify the pretrained model's filename instead of directory (you can download the zip in the release page).

upright2003 commented 1 year ago

您是指这个ckpt 文件吗? https://github.com/fishaudio/fish-diffusion/releases/tag/v1.11

cn-hubert-soft-600-singers-pretrained-v1.ckpt

leng-yue commented 1 year ago

是的

upright2003 commented 1 year ago

成功了 謝謝 但是發現一個問題就是 svc_cn_hubert_soft_finetune.py 無法產生 npy 文件 用 svc_hubert_soft.py 才能產生 npy 文件

89

leng-yue commented 1 year ago

额.. 你看看 wiki, 两个指定的训练文件夹不一样, 你要改一下.