torch<=2.8.0 torchaudio<=2.8.0 torchvision<=0.23.0 torchdata diffusers hydra-core omegaconf tqdm accelerate wandb einops transformers alias_free_torch h5py soundfile tensorboard swanlab fire sentencepiece librosa pypinyin g2p_en git+https://github.com/wenet-e2e/wespeaker.git moviepy av matplotlib