Describe the Bug

When following the [Metis TTS model instructions](https://github.com/open-mmlab/Amphion/tree/main/models/tts/metis) from the Amphion repository on a fresh Ubuntu 24.04 environment, the installation fails due to incompatible package dependencies. Additionally, attempting to run the example script results in runtime errors related to the package environment.

How To Reproduce

1. Clone the Amphion repo and navigate to models/tts/metis.

2. Create a new virtual environment and install dependencies listed in requirements.txt:

   pip install -r requirements.txt

3. Attempt to run the example inference script:

   python -m models.tts.metis.metis_infer_tse

4. Observe errors due to dependency conflicts and missing functionality.

Expected Behavior

The environment should install cleanly without package version conflicts, and running the inference script should execute successfully using the provided example input.

Screenshots / Logs

Example errors encountered:

Disabling PyTorch because PyTorch >= 2.1 is required but found 2.0.1
None of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
...
TypeError: 'xxx' object is not callable

Or other module import errors like:

ModuleNotFoundError: No module named 'pyopenjtalk'
ImportError: cannot import name 'LangSegment'

Environment Information

• Operating System: Ubuntu 24.04 LTS
• Python Version: Python 3.10.12
• CUDA Version: CUDA 12.5 (default on Ubuntu 24, not compatible with torch 2.0.1)
• Error Messages: See above

Additional Context

• torch==2.0.1 is incompatible with Ubuntu 24.04's default CUDA version.
• transformers==4.41.2 and accelerate==0.24.1 require PyTorch >= 2.1 for full functionality.
• Several packages (LangSegment, pyopenjtalk, phonemizer) require native dependencies or specific OS packages to compile correctly.
• Suggest providing a tested environment.yml or Dockerfile for reproducibility.