📌 Citation for KTTS Single Speaker Dataset

KTTS Single Speaker Dataset

📅 Published: 30 August 2024
📌 Version: 2
🔗 DOI: 10.17632/5c4dcvxdmb.2
👨‍💻 Contributors:

• Kh Mohmad Shafi
• Asif Ali Bhat
• Kamran Imtiyaz
• Javaid Iqbal

🗂 Original Dataset Link: Mendeley Data-KTTS Single Speaker Dataset

---

🔹 Enhanced & Cleaned Version

🔖 Omarrran/kashmiri_single_speaker_tts_hf_dataset_2024 is a cleaned, structured, and enhanced version of the original KTTS Single Speaker Dataset, modified for better accessibility and improved usability in Text-to-Speech (TTS) training.

📌 Improvements include:
✔ Strict audio-text alignment
✔ Corrected formatting issues
✔ Better dataset structure for machine learning applications

🚀 This version ensures improved reach and ease of use for researchers and developers working on Kashmiri TTS models.

Description This dataset has been primarily developed to facilitate the creation of text-to-speech systems for the Kashmiri language, a digitally underrepresented language predominantly spoken in the Jammu and Kashmir region of India. The dataset comprises 2,984 audio recordings in WAV format, each accompanied by its corresponding textual data in a separate file named ‘textcorpus.csv’. The ‘id’ column in the CSV file serves as a unique identifier, allowing users to efficiently locate the corresponding WAV files, which are systematically named according to the ‘id’ associated with the sentences they contain. All recordings feature a single male voice with a sample rate of 48,000 Hz, ensuring high-quality audio suitable for detailed phonetic analysis and machine learning applications. This consistent audio quality across the dataset provides a reliable foundation for training and testing text-to-speech models. Furthermore, the dataset can be a valuable resource for future research and development efforts aimed at enhancing digital accessibility for the Kashmiri-speaking population.

Omarrran/kashmiri_single_speaker_ttd_hf_dataset_2024

Dataset Description

This dataset contains Kashmiri text-to-speech (TTS) data, including paired text and audio files.

Dataset Statistics

• Total number of samples: 2984
• Number of samples in train split: 2387
• Number of samples in test split: 597

Audio Statistics

• Total audio duration: 15914.16 seconds (4.42 hours)
• Average audio duration: 5.33 seconds
• Minimum audio duration: 1.98 seconds
• Maximum audio duration: 12.60 seconds

Text Statistics

• Total number of words: 31365
• Average number of words per sample: 10.51
• Minimum number of words in a sample: 5
• Maximum number of words in a sample: 26

Dataset Structure

The dataset contains the following fields:

• file_name: Value(dtype='string', id=None)
• sentence: Value(dtype='string', id=None)
• audio: Audio(sampling_rate=16000, mono=True, decode=True, id=None)
• duration: Value(dtype='float64', id=None)
• sample_rate: Value(dtype='int64', id=None)
• __index_level_0__: Value(dtype='int64', id=None)

Usage Example

Here's a simple example of how to load and use this dataset:

from datasets import load_dataset

dataset = load_dataset('Omarrran/kashmiri_single_speaker_ttd_hf_dataset_2024')

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

# Get the first sample
first_sample = train_data[0]
print(f"Text: {first_sample['text']}")
print(f"Audio file: {first_sample['audio']}")

License

Please refer to the dataset card for license information.

Citation

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

@dataset{KTTS Single Speaker Dataset,
  author = {Shafi, Kh Mohmad; Bhat, Asif Ali; Imtiyaz, Kamran ; Iqbal, Javaid (2024), “KTTS Single Speaker Dataset”, Mendeley Data, V2, doi: 10.17632/5c4dcvxdmb.2
},
  title = {Omarrran/kashmiri_single_speaker_ttd_hf_dataset_2024},
  year = {2025},
  url = {https://huggingface.co/datasets/Omarrran/kashmiri_single_speaker_ttd_hf_dataset_2024}
  doi = {10.17632/5c4dcvxdmb.2 },
}