README.md

metadata

dataset_info:
  features:
    - name: id
      dtype: int64
    - name: raw_transcription
      dtype: string
    - name: transcription
      dtype: string
    - name: sentence_type
      dtype: string
    - name: speaker_id
      dtype: string
    - name: gender
      dtype: int64
    - name: audio
      struct:
        - name: array
          sequence: float32
        - name: path
          dtype: string
        - name: sampling_rate
          dtype: int64
  splits:
    - name: train
      num_bytes: 10662278090.528124
      num_examples: 9349
    - name: test
      num_bytes: 22809451.471875332
      num_examples: 20
  download_size: 4756967964
  dataset_size: 10685087542
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
license: apache-2.0
language:
  - ky

Description

Here is the English translation:

---

This dataset consists of approximately 7000 samples in Kyrgyz with 44100 sampling rate float32. The total speech duration is around 13 hours. The recording quality matches that of the studio. The dataset includes the following features:

• id: Sentence number.
• raw_transcription: Text of the sentence.
• transcription: Lowercase text, cleaned of punctuation marks.
• sentence_type: Sentence category (Declarative, Interrogative, Exclamatory).
• speaker_id: Speaker ID. Always value "1".
• gender: Gender of the speaker. Always "1".
• audio: Audio data, presented as a dictionary containing metadata such as sampling rate and audio in the form of a numpy array.

The speaker is Kanykey Bakashova.