metadata

language:
  - ta
size_categories:
  - 100K<n<1M
task_categories:
  - automatic-speech-recognition
pretty_name: Speech to Text (cleaned)
dataset_info:
  features:
    - name: path
      dtype: string
    - name: sentence
      dtype: string
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
  splits:
    - name: train
      num_bytes: 7336930447.304
      num_examples: 224581
    - name: validation
      num_bytes: 1796570819.462
      num_examples: 56146
    - name: test
      num_bytes: 1030720788.984
      num_examples: 31192
  download_size: 10119221124
  dataset_size: 10164222055.749998
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

Dataset Card for Dataset Name

Dataset Details

Dataset Description

This dataset is a combination of the Common Voice 16.0 and Open SLR datasets which is of 534 hours. It has been meticulously curated, normalized to a 16kHz sampling rate, and cleaned for better usability. This dataset aims to provide a comprehensive collection of speech data for various applications, including speech recognition, natural language processing, and machine learning research.

Curated by: Prajwal N. Pharande
Language: Tamil

Dataset Sources

Repository: https://commonvoice.mozilla.org/ta/datasets and https://www.openslr.org/127/

Uses

This dataset can be used for a wide range of applications, including:

Speech recognition system training and evaluation
Natural language processing tasks involving spoken language
Machine learning research on speech-related problems
Voice synthesis and voice cloning experiments

Dataset Structure

path : Name of audio file which is converted into array.
audio : Dictionary contaning path, array and sampling rate of an audio file.
sentence : Transcription of an audio file in Tamil language.

Dataset Creation

Data Collection and Processing

Audio converted to arrays : All audio samples have been normalized to a 16kHz sampling rate, ensuring consistency and high quality across the dataset.
Diverse Sources : The dataset combines data from Common Voice and Open SLR, providing a diverse range of voices, accents, and languages.
Cleaned Data : Extensive efforts have been made to clean the data, removing noise, punctuations, duplicates, and irrelevant metadata, to enhance usability and accuracy.

Who are the source data producers?

Mozilla : A large-scale, publicly available dataset of speech data collected by Mozilla, contributed by volunteers worldwide.
Open SLR : Various open speech and language resources collected and shared by the open-source community through Open Speech and Language Resources.

Dataset Card Authors

Prajwal N. Pharande

Dataset Card Contact

pharandeprajwal@gmail.com