nickthelegend/Telengana-Speech-Dataset

# 🎙️ Telangana Speech Dataset

A high-quality Telugu speech dataset from Telangana news broadcasts

Explore Dataset • Report Issues

---

📋 Dataset Overview

This dataset contains 501 audio segments of Telugu speech extracted from Telangana news broadcasts. Each segment includes professionally transcribed Telugu text, making it ideal for speech recognition, text-to-speech, and language model training tasks.

✨ Key Features

• Language: Telugu (Telangana dialect)
• Domain: News broadcasts and current affairs
• Total Samples: 501 audio clips
• Audio Format: MP3
• Duration Range: 1.04 - 7.66 seconds per clip
• Quality: Clean audio with professional transcriptions

---

📊 Dataset Structure

Each entry in the dataset contains:

Field	Type	Description
audio	Audio	Audio file in MP3 format
transcription	String	Telugu transcription (40+ characters)
caption	String	Duplicate of transcription
duration_ms	Integer	Audio duration in milliseconds
start_time_ms	Integer	Start timestamp of segment
end_time_ms	Integer	End timestamp of segment
id	String	Unique identifier
segment_index	Integer	Sequential segment number (0-500)
segment_filename	String	Original filename

---

🎯 Use Cases

This dataset is perfect for:

• 🗣️ Automatic Speech Recognition (ASR) - Training Telugu speech-to-text models
• 🔊 Text-to-Speech (TTS) - Building natural Telugu voice synthesis
• 🌐 Language Modeling - Understanding Telangana Telugu patterns
• 📰 News Analysis - Processing Telugu news content
• 🎓 Linguistic Research - Studying Telangana dialect characteristics

---

💾 Dataset Statistics

Total Audio Duration: ~42 minutes
Average Clip Length: 5.0 seconds
Transcription Length: 14-40 characters (Telugu script)
Total Segments: 501
Unique Source: Single news broadcast

---

🔧 Example Usage

Loading with Hugging Face Datasets

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("nickthelegend/TelenganaSpeechDataset")

# Access training split
train_data = dataset['train']

# Example: Get first sample
sample = train_data[0]
print(f"Transcription: {sample['transcription']}")
print(f"Duration: {sample['duration_ms']} ms")

# Play audio (if using IPython)
from IPython.display import Audio
Audio(sample['audio']['array'], rate=sample['audio']['sampling_rate'])

Processing Audio

import librosa

# Load audio from dataset
audio_array = train_data[0]['audio']['array']
sampling_rate = train_data[0]['audio']['sampling_rate']

# Extract features
mfcc = librosa.feature.mfcc(y=audio_array, sr=sampling_rate, n_mfcc=13)

---

📝 Sample Data

Transcription Example:

అందరికీ నమస్తే ఫస్ట్ హెడ్లైన్ చూడురి

Translation: "Hello everyone, let's see the first headline"

Context: News broadcast opening

---

🏷️ Topics Covered

The dataset includes segments on:

• Political news and governance
• Municipal elections
• Current affairs in Telangana
• Social issues
• Administrative updates

---

⚠️ Limitations

• Single source broadcast (limited speaker diversity)
• News domain specific vocabulary
• Telangana regional dialect (may not represent all Telugu variations)
• Background music in some segments (tagged as [సంగీతం])

---

📜 Citation

If you use this dataset in your research, please cite:

@dataset{telangana_speech_2024,
  title={Telangana Speech Dataset},
  author={nickthelegend},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/nickthelegend/TelenganaSpeechDataset}
}

---

🤝 Contributing

Found issues or have suggestions? Please open a discussion on the dataset page.

---

📧 Contact

For questions or collaboration opportunities, reach out through the Hugging Face dataset page.

---

Made with ❤️ for the Telugu NLP community

⭐ Star this dataset if you find it useful!