TTS-dataset-Manipur-latin

Dataset Description

This dataset comprises a collection of Manipuri (Romanized) speech audio recordings paired with their corresponding text transcriptions. It is designed to support research and development in Text-to-Speech (TTS) systems for the Manipuri language, specifically using Romanized script for text input.

Languages

This dataset is primarily in Manipuri (ISO 639-3: mni) and uses the Latin script for its text component.

Dataset Structure

The dataset is structured into a single train split.

Data Fields

Each entry in the dataset consists of two primary features:

• path: A string pointing to the absolute path of the audio file (in WAV format). When loaded into a Hugging Face Dataset object with the Audio feature type, this field enables direct audio playback.
• text: A string containing the Romanized Manipuri transcription of the corresponding audio.

Audio Details

All audio files are:

• Format: WAV
• Sampling Rate: 16000 Hz
• Channels: Mono

Usage

To load and use this dataset with the Hugging Face datasets library, you can use the following code:

from datasets import load_dataset

dataset = load_dataset("DayanandaThokchom/TTS-dataset-Manipur-latin")

# Access the train split
train_dataset = dataset["train"]

# Print the first example
print(train_dataset[0])

# Play an audio sample (requires IPython.display)
from IPython.display import Audio

# Assuming `train_dataset[0]["path"]` provides a dictionary with "array" and "sampling_rate" keys
# or points to a resolvable local path if using `Audio(decode=True)`
Audio(data=train_dataset[0]["path"]["array"], rate=train_dataset[0]["path"]["sampling_rate"])

Note: The audio paths in the vits_dataset.csv were originally extracted from embedded audio bytes, saved to temporary local WAV files, and then referenced in the CSV to ensure proper loading and preview functionality on the Hugging Face Hub.

Additional Information

• Language Details: The text is transcribed using the Latin script for Meitei language pronunciation.
• License: MIT License

Citation

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

@dataset{meitei_speech_dataset,
  author = {Dayananda Thokchom},
  title = {Meitei Speech Dataset},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/DayanandaThokchom/TTS-dataset-Manipur-latin}
}