search

myshell-ai / MeloTTS

High-quality multi-lingual text-to-speech library by MyShell.ai. Support English, Spanish, French, Chinese, Japanese and Korean.
MIT License
3.98k stars 476 forks source link

The pre-training model does not support Chinese. #69

Open v3ucn opened 3 months ago

v3ucn commented 3 months ago

Hi,Thank you for your open source project, but the pre-training model downloaded by default does not seem to support Chinese, and the trained model cannot produce Chinese voice.

shirubei commented 3 months ago

Same here.

using metadata like line below, 13000 steps but still failed to produce Chinese voice. It's just like noise or something .

processed_1.wav|Character1|ZH|好運不會在人家等候的那個地方自然來，而是經過彎彎曲曲，與困難的難以想像的道路才降臨的

shirubei commented 3 months ago

翻看其他帖子，#66 说底模是英文的，似乎训练不了中文模型

MujiKemp commented 3 months ago

It would be better if they provide more ckpts in PRETRAINED_MODELS in download_utils.py, then select based on language. Or a super big D/G/Dur for all languages...

anye1235 commented 2 months ago

训练代码异常： 安装readem里的训练数据格式 metadata.list的格式为：processed_1.wav|Character1|ZH|好運不會在人家等候的那個地方自然來，而是經過彎彎曲曲，與困難的難以想像的道路才降臨的

但是code里的 data_util.py _filter函数解析 _id, spk, language, text, phones, tone, word2ph = item；里面的phones 是否跳过的逻辑。导致解析错误，如果metadata.list里增加 phones, tone, word2ph空字符也是有问题，还请补充