- Test running CLI with minimal input file for each major supported model (e.g. at least 1 from each architecture) outputs expected files
- Test running CLI with pre-warmed cache directory containing model files does not repeat download
- Test running CLI with corrupt model files in cache directory throws expected error
- Test loading the separation class rather than using the CLI works as expected with all major supported models
- Test processing multiple files in a row outputs separate expected files
- Test processing file with multiple different models outputs separate expected files
- Test each of the architecure specific parameters works as expected in both CLI and class mode
- Generate oscillogram and spectrogram of model output for a short test file for each major supported model and compare to expected output to ensure separation is actually separating stems
- Add a few different test files with different properties, e.g. background noise, stems present, or genre of music and ensure separation works as expected for each
- Test that processing more than one distinct input file in sequence outputs separate files (not overwriting the first output with the second)
- Test that every model returned by list_models actually works
- Test each model architecture at least once using each supported Python version, to ensure we actually do legitimately support all of them!
- Test that loading a model of one architecture does NOT load the code from any other architecture