lilgoose777/tibetan-speech-english-text-dataset

Tibetan Speech Dataset with English Translations

Dataset Description

This dataset contains Tibetan speech recordings paired with transcriptions in Tibetan script and English translations. It is designed to support automatic speech recognition (ASR), machine translation, and text-to-speech (TTS) research for the Tibetan language, which is considered a low-resource language in NLP.

Supported Tasks

• —Automatic Speech Recognition (ASR): Train models to transcribe Tibetan speech to text
• —Speech Translation: Directly translate Tibetan speech to English text
• —Machine Translation: Translate between Tibetan and English text
• —Text-to-Speech (TTS): Generate natural Tibetan speech from text
• —Multilingual Speech Processing: Cross-lingual speech understanding

Languages

• —Tibetan (བོད་སྐད་): Primary language - ISO 639-1 code: bo
• —English: Translation language - ISO 639-1 code: en

Dataset Structure

Data Instances

Each example in the dataset contains:

{
  'audio': {
    'path': 'path/to/audio.wav',
    'array': array([...]),  # Audio waveform
    'sampling_rate': 16000
  },
  'sentence': 'English translation text',
  'tibetan': 'བོད་ཡིག་ཡིག་རིགས།'  # Tibetan script text
}

Data Fields

• —`audio` (Audio): Audio file containing Tibetan speech
• —Format: WAV
• —Sampling rate: 16,000 Hz
• —Channels: Mono

• —`sentence` (string): English translation of the Tibetan speech

• —`tibetan` (string): Transcription in Tibetan script (བོད་ཡིག་)

Data Splits

Currently, the dataset contains a single split:

Dataset Statistics

• —Total samples: 3921
• —Audio format: WAV, 16kHz, Mono
• —Average audio duration: [To be calculated]
• —Total audio duration: [To be calculated]
• —Script: Tibetan (Uchen script - དབུ་ཅན་)

Usage

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("lilgoose777/tibetan-speech-english-text-dataset")

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

print(f"Dataset size: {len(train_data)} examples")

Accessing Individual Examples

# Get first example
sample = dataset['train'][0]

# Access different fields
print("English:", sample['sentence'])
print("Tibetan:", sample['tibetan'])
print("Audio sampling rate:", sample['audio']['sampling_rate'])
print("Audio array shape:", sample['audio']['array'].shape)

Example: Play Audio

import IPython.display as ipd

# Play first audio sample
sample = dataset['train'][0]
ipd.Audio(sample['audio']['array'], rate=sample['audio']['sampling_rate'])

Example: ASR Training with Transformers

from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
from datasets import load_dataset

# Load dataset
dataset = load_dataset("lilgoose777/tibetan-speech-english-text-dataset")

# Load pretrained processor and model
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base")
model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base")

# Prepare audio
def prepare_dataset(batch):
    audio = batch["audio"]
    batch["input_values"] = processor(
        audio["array"], 
        sampling_rate=audio["sampling_rate"]
    ).input_values[0]
    batch["labels"] = processor.tokenizer(batch["tibetan"]).input_ids
    return batch

dataset = dataset.map(prepare_dataset, remove_columns=["audio"])

Example: Machine Translation

from datasets import load_dataset

dataset = load_dataset("lilgoose777/tibetan-speech-english-text-dataset")

# Access parallel text for translation
for example in dataset['train'][:5]:
    print(f"Tibetan: {example['tibetan']}")
    print(f"English: {example['sentence']}")
    print("-" * 50)

Dataset Creation

Source Data

[Describe where the audio recordings come from - e.g., audiobooks, interviews, read speech, etc.]

Data Collection

[Describe how the data was collected]

Speakers

• —Number of speakers: [To be added]
• —Gender distribution: [To be added]
• —Age range: [To be added]

Annotations

Annotation Process

• —Tibetan transcriptions: [Describe annotation process]
• —English translations: [Describe translation process]
• —Quality control: [Describe QC measures]

Annotators

[Information about who created the transcriptions and translations]

Considerations for Using the Data

Social Impact

This dataset aims to:

• —Support the preservation and digitalization of the Tibetan language
• —Enable development of speech technology for Tibetan speakers
• —Facilitate cross-lingual communication and understanding
• —Contribute to low-resource language research

Limitations

• —Limited speaker diversity
• —Domain-specific vocabulary [if applicable]
• —Dialectal variations may not be fully represented
• —Audio quality may vary across samples

Ethical Considerations

• —All audio recordings were obtained with appropriate permissions
• —Privacy considerations have been addressed
• —Cultural sensitivity has been maintained in translations

Additional Information

Dataset Curators

[Names and affiliations of dataset creators]

Licensing Information

This dataset is released under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

You are free to:

• —Share — copy and redistribute the material
• —Adapt — remix, transform, and build upon the material

Under the following terms:

• —Attribution — You must give appropriate credit

Citation Information

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

@dataset{tibetan_speech_dataset,
  author = lilgoose,
  title = {Tibetan Speech Dataset with English Translations},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/lilgoose777/tibetan-speech-english-text-dataset}
}

Contributions

Contributions to expand and improve this dataset are welcome! Please feel free to:

• —Report issues or errors
• —Suggest improvements
• —Add more data samples
• —Improve documentation

Contact: [Your contact information]

Acknowledgments

[Thank any organizations, funding sources, or individuals who contributed]

Related Datasets

• —[Link to other Tibetan language datasets if available]
• —[Link to similar low-resource speech datasets]

Changelog

• —v1.0.0 (2025-02-07): Initial release with 3921 samples