MyST / README.md
MagicLuke's picture
Update README.md
e7bbef0 verified
metadata
license: mit
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
  - config_name: finetune
    data_files:
      - split: train
        path: finetune/train-*
      - split: validation
        path: finetune/validation-*
      - split: test
        path: finetune/test-*
dataset_info:
  - config_name: default
    features:
      - name: file
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 16000
      - name: text
        dtype: string
      - name: duration
        dtype: float32
      - name: student_id
        dtype: string
      - name: date
        dtype: string
      - name: time
        dtype: string
      - name: module
        dtype: string
      - name: investigation
        dtype: string
      - name: part
        dtype: string
    splits:
      - name: train
        num_bytes: 43732745924.625
        num_examples: 181323
      - name: validation
        num_bytes: 5529107838.5
        num_examples: 23652
      - name: test
        num_bytes: 5385316354
        num_examples: 22592
    download_size: 50216196525
    dataset_size: 54647170117.125
  - config_name: finetune
    features:
      - name: file
        dtype: string
      - name: audio
        dtype:
          audio:
            sampling_rate: 16000
      - name: text
        dtype: string
      - name: duration
        dtype: float32
    splits:
      - name: train
        num_bytes: 12603945060.375
        num_examples: 51029
      - name: validation
        num_bytes: 2082115799.625
        num_examples: 8459
      - name: test
        num_bytes: 2224927204
        num_examples: 9200
    download_size: 16055315812
    dataset_size: 16910988064

NOTE: This is not a public dataset, I couldn't share access.

There are two config for this dataset:

  1. By default, is loading the raw dataset with basically everything of the dataset

  2. There is another configuration is for finetune use, which is a preprocessed myst subset with

    • cleaned transcription (in the format w2v2 output)
    • duration, only 2~20 sec audio entries will be seen in this subset
    • *local file name: only for reference use locally.
  from datasets import load_dataset

  myst = load_dataset("MagicLuke/MyST", split="test")
  myst = load_dataset("MagicLuke/MyST", "finetune", split="test")