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
4.56k stars 572 forks source link

How to train a new model #5

Open wxbool opened 7 months ago

wxbool commented 7 months ago

I am very interested in this project, when will the scripts for training new models be open? I would like to know how to train a new model.

Zengyi-Qin commented 7 months ago

There are many things on our teams' todo list. We will consider adding the training script if the model gets a lot of good feedback from the community

jsoncode commented 7 months ago

I’m also looking forward to the launch of the feature that can train new models

wxbool commented 7 months ago

Very much looking forward to being able to customize the training, this model works very well!

jurandfantom commented 7 months ago

count me in. Zengyi-Qin, this project is awesome thanks to pure fact it works as presented and quality is not only selected samples. Safe to say, community love your work and we all awaits any new feature that allows us use it beside of current set of voices.

Great job and happy to see people like you, contribute their projects for everybody to use <3

Zengyi-Qin commented 7 months ago

Hey guys - We have put the training code release in the todo list

SRbone commented 5 months ago

Hi, Zengqi-Qin and your team, thanks for your project. The model cards on HF (https://huggingface.co/myshell-ai/MeloTTS-Chinese/tree/main) is about 200 MB, but the model saved during my training is about 610 MB. There are some works to do, before infer? such as resize model or del some parameters?

pengpengtao commented 1 month ago

您好，Zengqi-Qin 和您的团队，感谢您的项目。 HF （https://huggingface.co/myshell-ai/MeloTTS-Chinese/tree/main） 上的模型卡约为 200 MB，但在我训练期间保存的模型约为 610 MB。在推断之前，还有一些工作要做？比如调整模型大小时还是一些参数？

请问你解决了吗，我也遇到一样的问题

pengpengtao commented 1 month ago

嘿伙计们 - 我们已经把训练代码发布放在了待办事项列表中

I have retrained the model and saved it, but the size of the model has become 600M. How can I modify it

drakitLiu commented 1 week ago

嘿伙计们 - 我们已经把训练代码发布到了待办事项列表中

我已经重新训练了模型并保存了，但是模型大小变成了600M，该如何修改

尝试将模型中的dict无关数据删除

drakitLiu commented 1 week ago

嗨，Zengqi-Qin 和你的团队，感谢你们的项目。HF 上的模型卡（https://huggingface.co/myshell-ai/MeloTTS-Chinese/tree/main）大约 200 MB，但我训练期间保存的模型大约 610 MB。在推断之前还有一些工作要做？例如调整模型大小或删除一些参数？

1、你可能保存了很多优化信息在模型中 2、你是在Linux上进行训练吗？我在widows上没有成功，可否告知您的训练环境呢？代码是否经过修改？

drakitLiu commented 1 week ago

1、你可能在模型中保存了很多优化信息 2、你是在Linux上进行训练吗？我在widows上修改没有成功，可否告知你的训练环境呢？代码是否经过？

1、你可能在模型中保存了很多优化信息 2、你是在Linux上进行训练吗？我在widows上修改没有成功，可否告知你的训练环境呢？代码是否经过？

pengpengtao commented 1 week ago

1、你可能在模型中保存了很多优化信息 2、你是在Linux上进行训练吗？我在widows上修改没有成功，可否告知你的训练环境呢？代码是否经过？

1、你可能在模型中保存了很多优化信息 2、你是在Linux上进行训练吗？我在widows上修改没有成功，可否告知你的训练环境呢？代码是否经过？

我训练结果也有问题，目前尝试从vits2开始训练，后面再加bert，最后再加tone等