ragunath-ravi/TamilVoiceCorpus

Tamil Conversational ASR Dataset

This is a dataset for Automatic Speech Recognition (ASR) focused on conversational Tamil, collected from various public sources on the web. Each sample is a short audio clip (averaging 10 seconds) paired with its corresponding transcription.

Dataset Summary

• Language: Tamil (ta)
• Domain: Conversational speech
• Average Duration per Clip: ~10 seconds
• Format: Audio (.wav) + text
• Sample Rate: 16kHz recommended
• Total Examples:
  • PureVox:
    • Train: 12,371
    • Test: 3,093
  • AmbientVox:
    • Train: 22,057
    • Test: 5,515

Dataset Structure

The dataset is organized into two subsets:

PureVox

The PureVox subset contains cleaner audio with minimal background noise, making it ideal for baseline ASR model training and evaluation.

Each example contains the following fields:

Field	Type	Description
file_name	string	Name of the original audio file
start_time	float	Start time of the segment in seconds
end_time	float	End time of the segment in seconds
duration	string	Duration of the segment
text	string	Ground truth transcription in Tamil
file_path	audio	The audio data itself
__index_level_0__	int64	Index identifier

AmbientVox

The AmbientVox subset contains conversational Tamil with realistic background noise and varying acoustic conditions, making it suitable for training robust ASR models for real-world applications.

Each example contains the following fields:

Field	Type	Description
duration	float64	Duration of the segment in seconds
file_name	string	Name of the original audio file
start_time	string	Start time of the segment
video_id	string	ID of the source video
text	string	Ground truth transcription in Tamil
end_time	string	End time of the segment
file_path	audio	The audio data itself

Usage

This dataset can be used for training or evaluating ASR models, especially those based on architectures like Whisper or Wav2Vec2.

You can load each subset individually:

from datasets import load_dataset

# Load PureVox subset (cleaner audio)
purevox_dataset = load_dataset("ragunath-ravi/TamilVoiceCorpus", name="PureVox")

# Load AmbientVox subset (with background noise)
ambientvox_dataset = load_dataset("ragunath-ravi/TamilVoiceCorpus", name="AmbientVox")

Comparison of Subsets

Feature	PureVox	AmbientVox
Background Noise	Minimal	Present
Size (Train)	12,371 examples	22,057 examples
Size (Test)	3,093 examples	5,515 examples
Total Dataset Size	~5.4 GB	~33 GB
Best Use Case	Baseline model training, Controlled evaluations	Robust model training, Real-world applications

Licensing

The dataset is compiled from public web sources and is available under the Apache 2.0 license. It is intended for research and educational purposes only. Ensure that you comply with the copyright terms of individual sources if redistributing or using commercially.

Citation

If you use this dataset in your research, please cite it appropriately or include the Hugging Face link to this dataset card.

Contributions

This dataset was curated and preprocessed by Ragunath-Ravi.