babysor / babysor/MockingBird

分享两个训练好的synthesizer模型

Open
#289 13 comments 10 reactions 0 assignees View on GitHub
documentation enhancement
Dominant language
Python
Stars
36.9k
Forks
5.2k
PR merge metrics
No merged PRs in 30d

Description

俩模型都是用最新的代码来训练的,不需要切换回0.0.1
第一个模型synthesizer-merged_110k,是在代码支持的四个数据集(aidatatang_200zh,magicdata,aishell3,data_aishell)上联合训练的。learning rate=0.001无衰减,batch size=128,iteration=110k。
第二个模型synthesizer-zhvoice_170k,是在[zhvoice](https://github.com/fighting41love/zhvoice)这个数据集上训练的。learning rate=0.001无衰减,batch size=128,iteration=170k。
两个模型我都已经测试过是可用的,不过第一个似乎比第二个好一点。我觉得问题是出在vocoder上,因为我现在用的vocoder(wavernn)并不是在zhvoice数据集上训练的,而我也懒得再训练一个vocoder。不过用hifigan的话效果倒是差不多,不过音色又有明显不同,挺有意思的。
关于训练,我这俩模型还是可以进一步优化的,现在第一个模型的loss是在0.24左右,第二个是在0.22左右,不过花的时间太久我就懒得继续训练了。
下面是下载链接:
[百度云](https://pan.baidu.com/s/1Gt2MQydfrreBi4htYhSUCQ) 密码: ir90
[Google Drive](https://drive.google.com/drive/folders/10LDxmZOto9ehPbZgTyvY2NzPEjHS4qdG?usp=sharing)

Contributor guide

No contributing guide indexed for this repository

Research direction

No files, tests, or code entry points are named. Start by reviewing the two model artifacts, their Baidu Cloud and Google Drive links, and the stated training details; clarify whether the intended outcome is to add, document, or maintain these models and what completion criteria apply.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
ai, machine-learning
Issue type
Documentation
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.