QingyuLiu0521/ICSD — reverse-engineered prompt

Build me a simple Python project for the ICSD infant cry and snoring detection dataset.

I want it to help someone download the dataset from Hugging Face with their token, preview a few demo audio clips, and run the provided baseline training or testing commands without having to dig through the code. It should make the workflow clear for a beginner, including setting up the environment, downloading the main dataset or the synthetic materials, training with synthetic data only, training with real plus synthetic data, and testing from a checkpoint.

Please keep it focused on the dataset and baseline sound event detection work, not a fancy product. Add helpful command line options, clear error messages, and a short usage guide. If the data or token is missing, explain what the user needs to do. Also include a simple way to run the pretrained embedding baseline after embeddings are extracted. Look up current Hugging Face docs online if needed.