metadata

dataset_info:
  config_name: es
  features:
    - name: input_values
      sequence: float32
    - name: input_length
      dtype: int64
    - name: labels
      sequence: int64
  splits:
    - name: train
      num_bytes: 28416160808
      num_examples: 91374
    - name: test
      num_bytes: 1946938848
      num_examples: 5286
  download_size: 30161672462
  dataset_size: 30363099656
configs:
  - config_name: es
    data_files:
      - split: train
        path: es/train-*
      - split: test
        path: es/test-*
license: gpl-3.0
language:
  - es
pretty_name: Common Voice 13.0 - Wav2Vec2 Preprocessed

Common Voice 13.0 - Wav2Vec2 Preprocessed

Basically took Common Voice 13.0, removed all languages but English and Spanish, removed all splits but train and test, then preprocessed data just as this tutorial for training Wav2Vec2 model for speech-recognition. Uploaded with push_to_hub function.

For now, just available in Spanish. Use as follows:

from datasets import load_dataset

train_ds = load_dataset("cristibp11/common_voice_13_0_wav2vec2_preprocessed", "es", split="train")