RAVE's generate script applies a model to large collections of audio files offline in batch mode

Beyond realtime use, the rave generate script applies a trained or exported RAVE model to batches of audio files. It accepts a model path (either a training run folder or a .ts scripted file), a list of input paths (files or directories), and an output directory. Each file is passed through the model (encode+decode) and written to the output folder. A streaming flag simulates streaming chunk-by-chunk processing for consistency with realtime behavior. This is useful for dataset augmentation, style-transfer preprocessing pipelines, or listening tests comparing model variants at scale.

Examples

rave generate model_path path_1 path_2 --out out_path, where path_X is a list of audio files or directories.

Assessment

You want to A/B test two RAVE configs by transforming 50 audio files through each. Which script handles this efficiently, and what flag ensures the output matches what the model would produce in nn~?