Super outdated DiffSinger training tutorial (Read the description)
Edit edit: This is extremely outdated, and inaccurate. Please consult the DiffSinger discord server to get assistance in training a DiffSinger model. I hate being directed to a discord too, but it's the best thing I can think of rn. I also deleted everything out of my google drive for personal reasons, so you'll have to find the checkpoints yourself. Theyre not hard to find via the diffsinger github repo
EDIT: conda doesn't come with git, apparently! Please enter the command "conda install -c anaconda git" before cloning the repo!
Feel free to ask questions! I'll be releasing the next part soon...ish?
Please note: This guide isn't meant to be professional, so I say silly things, make mistakes and use swear words.
Prerequisites:
The most basic knowledge of using a computer + python/conda
Have Anaconda3 installed (https://www.anaconda.com/download)
Have a training-ready DiffSinger dataset
If you do not have a dataset ready, don't fret! Here's a really in depth guide by Julieraptor to get you started!! https://docs.google.com/document/d/1z...
Another guide by Pix and SpoopyAce: https://docs.google.com/document/d/1u...
What we cover in this part:
Setting up the local environment with Conda + Git
Manually editing the acoustic config file
Binarizing and Training
Reading + Understanding Tensorboard
I'm really active in the DiffSinger discord server! There's tons of support there as well, feel free to join! / discord
Links (kinda in order):
DiffSinger Repo: https://github.com/openvpi/DiffSinger
nsf_hifigan from openVpi: https://github.com/openvpi/vocoders/r...
rmvpe model: https://drive.google.com/file/d/1TTrt...
torch command: https://pytorch.org/get-started/locally/
nnsvs-db-converter: https://github.com/UtaUtaUtau/nnsvs-d...
commands used:
Anything inside of curly brackets ({ }) will be a variable that YOU need to change.
install git: conda install -c anaconda git
create environment: conda create -n {name_of_env} python=3.8
activate env: conda activate {name_of_env}
clone diffsinger repo: git clone https://github.com/openvpi/DiffSinger...
binarization: python scripts/binarize.py --config {path_2_config.yaml}
training: python scripts/train.py --config {path_2_config.yaml} --exp_name {name_of_experiment} --reset
Chapters: (Contributed by @sharkP )
00:00 Intro and Requirements
01:52 Setting Up Conda Environment
03:21 Cloning Diffsinger Repository
04:09 Installing nsf_hifigan and rmvpe
05:26 Installing Torch and Diffsinger Requirements
07:08 Data Folder Setup
09:09 Setting Up The Config.yaml
20:54 Binarization / Preprocessing
23:15 Begin Training (use ctrl+c to stop training)
25:54 Tensorboard Setup and Explanations
32:00 Outro