Kailai Shen

config: { "train": { "log_interval": 200, "eval_interval": 1000, "seed": 2024, "epochs": 20000, "learning_rate": 2e-4, "betas": [0.8, 0.99], "eps": 1e-9, "batch_size": 32, "fp16_run": false, "lr_decay": 0.999875, "segment_size": 16384 , "init_lr_ratio": 1,...

In addition, I have also adopted the frontend processing method for both Chinese and English from GPT-SoVITS: https://github.com/RVC-Boss/GPT-SoVITS/tree/main/GPT_SoVITS/text.

However, it seems that the original MB-iSTFT-VITS2 (without any modifications) also exhibits this issue on LJSpeech. 

换成libritts下的cosyvoice.yaml就可以训练了