训练后为什么会变成这样？

训练完后收敛图不太正常，而且使用时也并不能生成输入的文本语音而是数据库的一段语音，我是按照教程里一步一步训练的，怎么解决？

No files or tests are named. Start by reproducing the tutorial training and checking the convergence plots, then trace the inference entry point to determine why input text returns a database audio clip; done means training converges normally and arbitrary input text produces the expected synthesized speech.

Written by the indexing model from the issue text.