README.md

metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: label
      dtype:
        class_label:
          names:
            '0': down
            '1': 'on'
  splits:
    - name: train
      num_bytes: 117237120.74
      num_examples: 3706
    - name: validation
      num_bytes: 16469406
      num_examples: 521
    - name: test
      num_bytes: 15996007
      num_examples: 499
  download_size: 135214158
  dataset_size: 149702533.74

Dataset Card for "down_on"

This is a demo dataset for acoustic classification, consisting of utterances of down and on extracted from the superb ks dataset.

down_on_hub.py illustrates the methodology for creating such a dataset from wav files. down_on_copy.py was used to write the wav files. down_on_create.py is the more direct method.

See the model MatsRooth/wav2vec2-base_down_on for the demo application.