0x0：实验 序号 teacher layer 底模 数据增强 训练集 时长 01 contentvec 12 contentvec Default WenetSpeech4TTS 375h 02 contentvec 12 contentvec Default WenetSpeech4TTS + m4singer + opensinger + opencpop + popcs 375h 03 chinese-hubert-large 12 None Default WenetSpeech4TTS + m4singer + opensinger + opencpop + popcs 375h 04 chinese-hubert-large 12 None + Noise WenetSpeech4TTS + m4singer + opensinger + opencpop + popcs 375h 05 contentvec 12 contentvec + Noise WenetSpeech4TTS + m4singer + opensinger + opencpop + popcs 375h 06 contentvec 12 None + Noise WenetSpeech4TTS + m4singer + opensinger + opencpop + popcs 375h 0x1：总结 loss可以收敛，效果还行 loss可以收敛，效果不错 loss可以收敛，效果不行 loss可以收敛，效果未知 loss不能收敛 loss可以收敛，效果未知