自己训练的模型使用不了

我想自己训练个模型，随便找了20句话左右扔进去训练。loss已经是0.2左右了，但是用的时候并不能生成输入的文本语音，而是随机生成一段训练集里面的音频。这是为什么呢？
是因为我的训练集太小了不能收敛吗，我是小白，不是很懂。

Start by reproducing the reported training and inference setup with the approximately 20 training utterances described in the issue, then inspect the model's training and inference configuration. The investigation is complete when it identifies whether the small dataset or another setup problem causes generated speech to select random training audio instead of matching the input text.

Written by the indexing model from the issue text.