v1.2 vocal quality improves greatly, but audio reference degrades ?

[opencvbaby]

audio reference not as effective as v1 ? why ?

[opencvbaby]

我的意思是原来的模型，参考音频时，生成的歌曲风格和参考音频非常相似，而v1.2模型参考音频生成的歌曲风格几乎和参考音频没什么关系。这是为何？

My point is that with the original model, the songs generated while referencing the audio have a style that is very similar to the reference audio. However, with the v1.2 model, the style of the songs generated from the reference audio is nearly unrelated to the reference audio. Why is that?