README.md

metadata

dataset_info:
  - config_name: audiobooks
    features:
      - name: audio
        dtype: audio
      - name: text
        dtype: string
      - name: duration
        dtype: float64
      - name: file_id
        dtype: string
      - name: speaker_id
        dtype: string
    splits:
      - name: train
      - name: dev
      - name: test
  - config_name: crowdsourced_web
    features:
      - name: audio
        dtype: audio
      - name: text
        dtype: string
      - name: duration
        dtype: float64
      - name: file_id
        dtype: string
      - name: speaker_id
        dtype: string
    splits:
      - name: train
      - name: dev
      - name: test
  - config_name: crowdsourced_tg
    features:
      - name: audio
        dtype: audio
      - name: text
        dtype: string
      - name: duration
        dtype: float64
      - name: file_id
        dtype: string
      - name: speaker_id
        dtype: string
    splits:
      - name: train
configs:
  - config_name: audiobooks
    data_files:
      - split: train
        path: audiobooks/train.parquet
      - split: dev
        path: audiobooks/dev.parquet
      - split: test
        path: audiobooks/test.parquet
  - config_name: crowdsourced_web
    data_files:
      - split: train
        path: crowdsourced_web/train.parquet
      - split: dev
        path: crowdsourced_web/dev.parquet
      - split: test
        path: crowdsourced_web/test.parquet
  - config_name: crowdsourced_tg
    data_files:
      - split: train
        path: crowdsourced_tg/train.parquet
license: cc-by-4.0
task_categories:
  - automatic-speech-recognition
language:
  - tt
tags:
  - Tatar
  - ASR
  - Speech
  - Audio
  - Turkic
  - STT
pretty_name: Tatar Speech Corpus ASR
size_categories:
  - 100K<n<1M

Tatar Speech Corpus ASR

[Original repository] [Original site]

Dataset Summary

Tatar Speech Corpus ASR is a speech dataset sourced from this GitHub repository. TatSC contains 269.1 hours of transcribed speech with 271,914 utterances. It is the first open-source Tatar speech corpus covering both crowdsourced and audiobooks data.

Dataset Structure

Parts:
The dataset contains a single set of recordings, though they come from two different sources:

• Crowdsourced recordings (thru Telegram bot and website)
• Aligned audiobooks

Data Fields:

• audio: The speech audio file in the Tatar language
• text: The transcription of the audio file in Tatar
• duration: Length of the audio in seconds
• file_id: Unique identifier for the audio file
• speaker_id: Identifier for the speaker in the recording

Data Processing

Text Processing:
Information about text normalization or preprocessing is not explicitly provided in the source.

Audio Processing:
The audio samples range from very short utterances (0.21s) to longer segments (31.1s).

Technical Details:

Dataset Type	speech corpus for ASR
Language	tt, Tatar
Speech Style	mixed (crowdsourced and audiobooks)
Content	varied
Audio Parameters	16 kHz sampling rate
File Format	MP3, TXT (UTF-8)

Durations:

Audiobooks	23.02 hours
Crowdsourced (Telegram)	143.36 hours
Crowdsourced (Web)	97.29 hours
Total duration	269.1 hours

License

This dataset is distributed under the CC BY 4.0 license.