是否实验过m2_hat用linear频谱计算误差呢？

您好，有注意到用了postnet处理mel为linear，但是看训练过程，并没有用linear的频谱，而是用mel的频谱。

问下是否实验过m2_hat用linear频谱计算误差呢？
如果有实验过，能否谈一下结果如何呢？

下面贴上loss的计算：
# Backward pass
m1_loss = F.mse_loss(m1_hat, mels) + F.l1_loss(m1_hat, mels)
m2_loss = F.mse_loss(m2_hat, mels)
stop_loss = F.binary_cross_entropy(stop_pred, stop)

The issue provides only the loss snippet, including m1_hat, m2_hat, mels, and stop_pred; no file or test is named. Start by locating the training loss implementation and determine whether m2_hat can be compared with a linear spectrum instead of mels. Done would require an agreed experiment and documented results.

Written by the indexing model from the issue text.