a CustomTkInter GUI for processing and training DiffSinger models

DiffTrainer brings together the most useful tools for DiffSinger in one easy, graphical package.

• corpus_segmenter to shorten long wav/lab pairs
• liteconvert for converting wav+lab data to wav/csv
• SOME for estimating pitch
• DiffSinger's primary training
• OpenUtau export scripts

News for existing users

Installation:

• Easy Mode(Windows):
  • make sure a compatible version of CUDA Toolkit is installed
    • compatible versions: 11.8, 12.1, 12.4, 12.6, 12.8, 12.9
  • install miniconda (full conda works but comes with tons of stuff not needed, waste of space)
  • reconfigure_conda_env.bat
  • run_gui.bat
  • Update Tools button to finish
  • in future usage, run_gui.bat will check for DiffTrainer updates before launching
• Easy-ish Mode(Linux)
  • make sure a compatible version of CUDA Toolkit is installed
    • compatible versions: 11.8, 12.1, 12.4, 12.6, 12.8, 12.9
  • install miniconda (full conda works but comes with tons of stuff not needed, waste of space)
  • conda env create --file assets/environment_LINUX.yml
  • always activate newly created difftrainer environment before running any python files
  • run auto_torch.py
  • run difftrainer.py
  • Update Tools button to finish
  • in future usage, check_update.py checks for DiffTrainer updates before running difftrainer.py
• Normal Mode(Windows/Linux)
  • make sure a compatible version of CUDA Toolkit is installed
    • compatible versions: 11.8, 12.1, 12.4, 12.6, 12.8, 12.9
  • bring your environment manager of choice(or don't, if you live dangerously)
  • python 3.10
  • install requirements.txt (less tested than the conda equivalent assets/environment.yml)
  • always activate your environment before running any python files
  • run auto_torch.py OR add a compatible version of PyTorch yourself (>=2.4, <=2.8, CUDA enabled)
  • run difftrainer.py
  • Update Tools button to finish
  • in future usage, check_update.py checks for DiffTrainer updates before running difftrainer.py
• ??? Mode(Mac)
  • see Linux instructions, but use the non-Linux environment.yml for conda
  • auto_torch should install Torch 2.11 for M-series, 2.8 for anything else
  • honestly you're on your own in general but in theory it's supposed to run
  • I'm willing to try to make it work better but my tester Mac is Intel(a lost cause)

Known issues:

• (Windows)Pytorch fails to extract during environment setup
  • try enabling long paths
• langloader editor sometimes hides behind main window
• if you type in the save interval or batch size boxes, an error appears in the terminal window
  • no actual impact, just enter your number and ignore it
• do not name checkpoint folders just "acoustic" or "variance", it conflicts with the onnx export cleanup

on environment management and .bats

Starting from DiffTrainer 0.4, environments will be once again self-manageable. All .py files will act using the environment that is active when the file is run. However, all .bat files will still be written to activate a Conda environment(specifically named difftrainer) and run the file from that environment. If you want to use a different environment manager or different environment name, you will need to run the .py files directly from the command line or write your own automated activation scripts. No .bats are strictly required, they are purely for convenience.

1. reconfigure_conda_env.bat:
   • Attempts to locate conda at either the custom location installed with the old conda_installer.bat or one of the default install locations. If it is not found at any of these locations, it will still attempt to call Conda with the system PATH.
   • Attempts to remove the difftrainer environment from the identified Conda. Continues if not found.
   • Recreates the difftrainer environment under the identified Conda
   • Runs auto_torch.py within the new difftrainer environment to finish environment setup
2. run_gui.bat:
   • Attempts to locate conda at either the custom location installed with the old conda_installer.bat or one of the default install locations. If it is not found at any of these locations, it will still attempt to call Conda with the system PATH.
   • Activates the difftrainer environment before running check_update.py(which then runs difftrainer.py)
3. launch_tensorboard.bat:
   • Attempts to locate conda at either the custom location installed with the old conda_installer.bat or one of the default install locations. If it is not found at any of these locations, it will still attempt to call Conda with the system PATH.
   • Activates the difftrainer environment before running Tensorboard pointed at the main checkpoints folder.
4. run_quickinference.bat:
   • Attempts to locate conda at either the custom location installed with the old conda_installer.bat or one of the default install locations. If it is not found at any of these locations, it will still attempt to call Conda with the system PATH.
   • Activates the difftrainer environment before running quickinference.py.