zmh56/SDD-for-DMPF-MPSC ? reverse-engineered prompt

Build me a Python project that does speech based depression detection using five perspectives: voiceprint, emotion, pause, energy, and tremor.

I want the repo to let me run each feature extraction step separately, then combine everything in a fusion training step for final prediction. Please make the scripts easy to configure with audio paths, segment length, and other settings, and keep the workflow clear for someone who just wants to point it at a dataset and run it.

If anything is unclear, look up the current docs online or make sensible defaults so the whole pipeline still works. I also want a clean way to run the 5 fold training and a simple explanation of what each step outputs, so I can go from raw audio to extracted features to fused depression detection without having to figure out the internals first.