Steps to run it on wsl + rocm [not issue is a howto]

My setup is an 13600k, 32gb ram, rx7800xt, windows 10, wsl2 (ubuntu 24.04)

rocm7.2.1
python 3.12
torch-2.9.1
triton-3.5.1
torchaudio-2.9.0

torchcodec dont work with rocm so need to monkeypatch (app.py in this case) on the beginning by adding

import torchaudio

def _fake_load_with_torchcodec(path, *args, **kwargs):
    return torchaudio.load(path, *args, **kwargs)

def _fake_save_with_torchcodec(path, tensor, sample_rate, *args, **kwargs):
    return torchaudio.save(path, tensor, sample_rate, *args, **kwargs)

torchaudio.load_with_torchcodec = _fake_load_with_torchcodec
torchaudio.save_with_torchcodec = _fake_save_with_torchcodec

also i had to set optimize to False in self.voxcpm_model = voxcpm.VoxCPM(voxcpm_model_path=model_dir, optimize=True)

Start with the issue's WSL2, ROCm, Python, PyTorch, and torchaudio setup details, then inspect app.py and the VoxCPM initialization mentioned in the report. Document the required compatibility workarounds and confirm that the documented steps let a WSL2 user run the project with this setup.