nikhilsaipagidimarri/Chaganti_Voice_with_CC_Dataset

Dataset Card for Chaganti Voice with CC Dataset

A curated Telugu speech dataset of Sri Chaganti Koteswara Rao's spiritual discourses (pravachanams), paired with auto-generated transcriptions. Built for training or fine-tuning ASR and TTS / Voice Cloning models in the Telugu language.

Dataset Details

Dataset Description

This 4.36 GB dataset contains high-quality audio clips of Sri Chaganti Koteswara Rao's Telugu spiritual discourses, precisely chunked and aligned with auto-generated YouTube transcriptions. Silence has been trimmed from each clip for optimized ML training pipelines.

• Curated by: nikhilsaipagidimarri
• Funded by: N/A
• Shared by: nikhilsaipagidimarri
• Language(s) (NLP): Telugu (te)
• License: CC BY-NC 4.0
• Total Audio Duration: +5 hours
• Total Size: 4.36 GB (778 instances)

Dataset Sources

• Repository: nikhilsaipagidimarri/Chaganti_Voice_with_CC_Dataset
• Paper: N/A
• Demo: N/A

Uses

Direct Use

This dataset is intended for:

• Training and fine-tuning Telugu ASR (Automatic Speech Recognition) models
• Training and fine-tuning Telugu TTS (Text-to-Speech) models
• Voice cloning research using a single high-quality Telugu speaker
• Academic research on low-resource Telugu speech

Out-of-Scope Use

• Commercial use of any kind — prohibited under CC BY-NC 4.0
• Multi-speaker speech tasks — this dataset contains a single speaker only
• Speaker diarization or speaker verification tasks
• Any use that violates YouTube's Terms of Service

Dataset Structure

This dataset has been fully optimized into Parquet shards. This means you do not need to download raw .wav files manually; the Hugging Face datasets library will automatically handle the audio decoding and convert the files into numpy arrays for immediate model training.

File Layout

data/
  train-00000-of-00020.parquet
  train-00001-of-00020.parquet
  train-00002-of-00020.parquet
  train-00003-of-00020.parquet
  train-00004-of-00020.parquet
  train-00005-of-00020.parquet
  train-00006-of-00020.parquet
  train-00007-of-00020.parquet
  train-00008-of-00020.parquet
  ...
  train-00019-of-00020.parquet

Data Instance

Each loaded instance contains:

• audio — Dictionary containing the raw audio waveform array, sampling rate, and reference path.
• transcription — Telugu text of what is spoken in the clip.

Example:

{
  'audio': {
      'path': 'utt_001.wav', 
      'array': array([ 0.0000000e+00, -3.0517578e-05, ...]), 
      'sampling_rate': 16000
  },
  'transcription': 'దశావతారాలలో అత్యంత ప్రముఖమైనటువంటి...'
}

Loading the Dataset

Because of the Parquet optimization, you can stream or load this dataset instantly in Python:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("nikhilsaipagidimarri/Chaganti_Voice_with_CC_Dataset", split="train")

# Access the first sample
sample = dataset[0]
print("Text:", sample["transcription"])
print("Audio Array:", sample["audio"]["array"])

Dataset Creation

Curation Rationale

There is a significant lack of high-quality, publicly available Telugu speech datasets — especially from a consistent, clear single speaker. Sri Chaganti Koteswara Rao's discourses are widely available on YouTube with auto-generated captions, making them a strong candidate for building an aligned Telugu speech corpus for ASR/TTS research.

Source Data

Data Collection and Processing

1. Audio was collected from publicly available YouTube videos of Sri Chaganti Koteswara Rao
2. Transcriptions were sourced from YouTube's auto-generated Telugu captions
3. Audio was precisely chunked to align with transcript segments (~25–40 seconds per clip)
4. Silence was trimmed from each clip using automated VAD (Voice Activity Detection)

Total audio duration across all clips: +5 hours

Source Videos:

#	YouTube URL
1	https://youtu.be/3G1lu0h7vz8
2	https://youtu.be/MAeew4uIHs4
3	https://youtu.be/4h1R32q3bC8
4	https://youtu.be/TEO3wNtQswk
5	https://youtu.be/iz_OilT9hAA
6	https://youtu.be/pdwNwdxqbcI

Who are the source data producers?

The original audio content was created and delivered by Sri Chaganti Koteswara Rao, a renowned Telugu scholar and spiritual speaker. The videos are published by Sri Chaganti Media on YouTube.

Annotations

Annotation process

Transcriptions were obtained from YouTube's auto-generated Telugu caption system and aligned to audio chunks programmatically. No manual correction has been applied.

Who are the annotators?

Auto-generated by YouTube's speech recognition system. No human annotators were involved.

Personal and Sensitive Information

The dataset contains the voice and speech of a public figure (Sri Chaganti Koteswara Rao) delivering public spiritual discourses. No private, sensitive, or personally identifiable information is present beyond the speaker's publicly shared content.

Bias, Risks, and Limitations

• Transcriptions are auto-generated and may contain errors in Telugu text, especially for Sanskrit shlokas and proper nouns
• All audio is from a single male speaker — models trained on this data will not generalize to other speakers or genders
• Dataset contains approximately +5 hours of audio (778 utterances) — highly suitable for fine-tuning but not for training large-scale models entirely from scratch
• Speaker accent and vocabulary are domain-specific (spiritual/religious discourse)

Recommendations

• This dataset is best used for fine-tuning pre-trained models
• Transcription quality should be verified before use in high-accuracy ASR pipelines
• Models trained on this data should not be used for commercial voice synthesis or impersonation

Citation

BibTeX:

@dataset{chaganti_voice_cc_2026,
  author    = {nikhilsaipagidimarri},
  title     = {Chaganti Voice with CC Dataset},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/nikhilsaipagidimarri/Chaganti_Voice_with_CC_Dataset},
  license   = {CC BY-NC 4.0}
}

APA:

nikhilsaipagidimarri. (2026). Chaganti Voice with CC Dataset [Dataset]. Hugging Face. https://huggingface.co/datasets/nikhilsaipagidimarri/Chaganti_Voice_with_CC_Dataset

Glossary

• ASR — Automatic Speech Recognition: converting spoken audio to text
• TTS — Text-to-Speech: converting text to spoken audio
• Pravachanam — A Telugu/Sanskrit term for a spiritual discourse or lecture
• VAD — Voice Activity Detection: automated removal of silence from audio

More Information

For questions, issues, or contributions, please open a discussion on the dataset repository.

Dataset Card Authors

nikhilsaipagidimarri

Dataset Card Contact

Reach out via the Hugging Face discussion tab on this dataset's repository page.

Acknowledgements

All credit for the original discourses goes to Sri Chaganti Koteswara Rao and Sri Chaganti Media for making these spiritual teachings publicly available on YouTube.