metadata

dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: text
      dtype: string
    - name: normalized_text
      dtype: string
    - name: speaker
      dtype: string
  splits:
    - name: test
      num_bytes: 7993646
      num_examples: 95
    - name: train
      num_bytes: 68421127
      num_examples: 277
  download_size: 76394317
  dataset_size: 76414773
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*
      - split: train
        path: data/train-*
language:
  - hr
pretty_name: Mići Princ
size_categories:
  - n<1K

Mići Princ - Chakavian ASR Dataset

Antoine de Saint-Exupéry's The Little Prince, translated into Chakavian dialect of Croatian and read by native speakers. Full dataset (with all metadata and structure) is published at http://hdl.handle.net/11356/1765.

Dataset composition

Attributes:

audio: audio chunk. In train, the audio can include multiple speakers and has a duration in [0, 30] seconds for easier ASR training. In test, only one speaker is present in an instance.
text: Text with punctuation, newlines, bullets, and special characters.
normalized_text: Text with some punctuation removed and special characters remapped for easier training. Characters, not present in standard Croatian (ȅ,é,ȋ,ȉ,ȃ,ȁ), are substituted with non-special characters (e,i,a). Bullets, newlines, and quotation marks are removed.
speaker: In train, this attribute is always an empty string. In test it accurately represents the speaker of the chunk.

Extent:

train: 277 instances
test: 95 instances

Authors:

Nikola Ljubešić https://huggingface.co./nljubesi
Peter Rupnik https://huggingface.co./5roop
Tea Perinčić