Packed discrete-unit files for SpeechGR experiments on SLUE-SQA-5.
The units were produced with HuBERT layer 22 and a K=500 k-means model, then deduplicated with consecutive counts retained. The packed format avoids one .code and .cnt file per utterance.
Files
documents.npz: packed document/passage units
train.npz: packed train question units
validation.npz: packed validation question units
test.npz: packed test question… See the full description on the dataset page: https://huggingface.co/datasets/dodofk/slue-sqa-code-l22-c500.
