search

daniilrobnikov / vits2

VITS2: Improving Quality and Efficiency of Single-Stage Text-to-Speech with Adversarial Learning and Architecture Design
https://vits-2.github.io/demo/
MIT License
461 stars 47 forks source link

AssertionError: min value is nan #3

Open wlz987 opened 1 year ago

wlz987 commented 1 year ago

Traceback (most recent call last): File "/home/work/miniconda3/envs/vits2/lib/python3.11/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap fn(i, *args) File "/home/work/语音相关/文字转语音/vits2/train_ms.py", line 92, in run train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler, [train_loader, eval_loader], logger, [writer, writer_eval]) File "/home/work/语音相关/文字转语音/vits2/train_ms.py", line 131, in train_and_evaluate y_hat_mel = wav_to_mel(y_hat.squeeze(1), hps.evaluate.n_fft, hps.evaluate.n_mels, hps.data.sample_rate, hps.evaluate.hop_length, hps.evaluate.win_length, hps.evaluate.f_min, hps.evaluate.f_max) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/work/语音相关/文字转语音/vits2/utils/mel_processing.py", line 64, in wav_to_mel assert torch.min(y) >= -1.0, f"min value is {torch.min(y)}" ^^^^^^^^^^^^^^^^^^^^ AssertionError: min value is nan

/home/work/语音相关/文字转语音/vits2/datasets/custom_base/filelists/train.txt:的一条数据 /home/work/语音相关/音频数据/Genshin_zh/barbara/vo_ABDLQ001_3_barbara_01.wav|0|yua˧˥n yua˧˥n la˧˥i ni˧ za˨˩˦i zhe˨˩˦ li˧ , luo˧˥ sha˥ li˨˩˦ ya˨˩˦ xia˧o jie˧ , cha˨˩˦nɡ shi˥ ba˥n de huo˧˥ do˨˩˦nɡ ma˧ sha˨˩˦nɡ jiu˨˩˦ …

这条数据的原文: /home/work/语音相关/音频数据/Genshin_zh/barbara/vo_ABDLQ001_3_barbara_01.wav|barbara|原、原来你在这里,罗莎莉亚小姐,唱诗班的活动马上就…

wlz987 commented 1 year ago

˧,˨˩˦是声调,拼音的四个声调

daniilrobnikov commented 1 year ago

Hi @longzewei!

I seems that the file /Genshin_zh/barbara/vo_ABDLQ001_3_barbara_01.wav is out of range from [-1 to 1]. One probable answer is that the file is corrupted

Can you send me this file for further assessment?

wlz987 commented 1 year ago

你这拉的新的代码就没有这个问题了

wlz987 commented 1 year ago

应该是 你改了 preprocess/mel_process.py 原来是 preprocess/aul_mel.py 还是什么的, 也没太仔细看

daniilrobnikov commented 1 year ago

Yes, recently, I refactored the code, so the naming of some files have changed as well

You should try to use:

cd vits2 # make sure you are in the project root
python preprocess/mel_transform.py --data_dir /path/to/custom_dataset -c datasets/custom_base/config.yaml