lilgoose777/tibetan-audio-english-datasets-full

Tibetan Audio-English Translation Dataset (Full)

A comprehensive dataset containing 17,278 audio-text pairs of Tibetan speech with English translations, created by merging and standardizing 6 high-quality Tibetan audio datasets.

📊 Dataset Overview

• Total Samples: 17,278
• Audio Format: WAV, 16kHz sampling rate
• Languages: Tibetan (audio) → English (text)
• Task Types: Automatic Speech Recognition (ASR), Translation, Text-to-Speech (TTS)
• Average Audio Duration: Varies by sample (typically 2-10 seconds)

🗂️ Source Datasets

This dataset is a carefully merged and standardized collection from 6 source datasets:

Dataset	Samples	Source
tibetan-speech-english-text-dataset	3,921	lilgoose777/tibetan-speech-english-text-dataset
merged-tibetan-titung-goose	5,188	lilgoose777/merged-tibetan-titung-goose
tibetan-to-english-audio-dataset	1,178	Titung/tibetan-to-english-audio-dataset
tibetan_plant_audio_casted	294	milanakdj/tibetan_plant_audio_casted
tibetan_plant_audio_casted_augmented	294	milanakdj/tibetan_plant_audio_casted_augmented
tibetan-speech-english-text-dataset-new-updated	6,403	lilgoose777/tibetan-speech-english-text-dataset-new-updated

All datasets have been:

• ✅ Standardized to a uniform format
• ✅ Validated for audio quality
• ✅ Cleaned for missing or invalid entries
• ✅ Resampled to 16kHz
• ✅ Shuffled for better distribution

📋 Dataset Structure

Data Fields

{
    'audio': {
        'path': str,              # Original audio file path
        'array': np.ndarray,      # Audio waveform as numpy array
        'sampling_rate': int      # Sampling rate (16000 Hz)
    },
    'sentence': str               # English translation/transcription
}

Data Splits

Split	Samples
train	17,278

Note: Currently provided as a single training split. Users can create their own train/validation/test splits as needed.

🚀 Usage

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("lilgoose777/tibetan-audio-english-datasets-full")

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

print(f"Total samples: {len(train_data)}")

Accessing Audio and Text

# Get a single sample
sample = train_data[0]

# Access audio
audio = sample['audio']
audio_array = audio['array']      # numpy array of audio waveform
sampling_rate = audio['sampling_rate']  # 16000 Hz

# Access text
sentence = sample['sentence']

print(f"Sentence: {sentence}")
print(f"Audio shape: {audio_array.shape}")
print(f"Duration: {len(audio_array) / sampling_rate:.2f} seconds")

Playing Audio (in Jupyter/Colab)

import IPython.display as ipd

# Play the audio
sample = train_data[0]
ipd.Audio(sample['audio']['array'], rate=sample['audio']['sampling_rate'])

Creating Train/Validation Split

# Split into train (90%) and validation (10%)
dataset = dataset['train'].train_test_split(test_size=0.1, seed=42)

train_dataset = dataset['train']      # 15,550 samples
validation_dataset = dataset['test']  # 1,728 samples

print(f"Train: {len(train_dataset)}")
print(f"Validation: {len(validation_dataset)}")

Streaming for Large-Scale Processing

# Load in streaming mode to avoid downloading everything at once
dataset = load_dataset(
    "lilgoose777/tibetan-audio-english-datasets-full", 
    split="train",
    streaming=True
)

# Iterate through samples
for sample in dataset.take(5):
    print(sample['sentence'])

🎯 Use Cases

This dataset is suitable for:

1. Automatic Speech Recognition (ASR)

   • Training Tibetan speech recognition models
   • Fine-tuning multilingual ASR models (e.g., Whisper, Wav2Vec2)

2. Machine Translation

   • Tibetan-to-English translation models
   • Speech-to-text translation systems

3. Text-to-Speech (TTS)

   • Training Tibetan voice synthesis models
   • Creating multilingual TTS systems

4. Low-Resource Language Research

   • Studying endangered language preservation
   • Developing tools for language documentation

5. Cross-Lingual Studies

   • Tibetan-English linguistic analysis
   • Phonetic and prosodic research

🔧 Preprocessing Applied

The following preprocessing steps were applied during dataset creation:

1. Standardization: All source datasets were converted to a uniform schema with audio and sentence columns
2. Audio Normalization: All audio resampled to 16kHz mono
3. Quality Filtering: Samples with missing audio or empty text were removed
4. Shuffling: Dataset shuffled with seed=42 for randomization
5. Validation: Each sample validated for audio integrity and text encoding

⚠️ Limitations

• Language Variety: Dataset may not cover all Tibetan dialects uniformly
• Domain: Content distribution depends on source datasets (may be skewed toward certain topics)
• Audio Quality: Variable recording quality across different source datasets
• Single Split: No predefined train/validation/test splits (users must create their own)
• Translation Quality: English translations may vary in accuracy and style

📝 Citation

If you use this dataset in your research, please cite the original source datasets:

@dataset{tibetan_audio_english_full,
  title={Tibetan Audio-English Translation Dataset (Full)},
  author={lilgoose777},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/lilgoose777/tibetan-audio-english-datasets-full}
}

🙏 Acknowledgments

Special thanks to the creators and contributors of the original datasets:

• lilgoose777 - tibetan-speech-english-text-dataset series
• Titung - tibetan-to-english-audio-dataset
• milanakdj - tibetan_plant_audio datasets

This merged dataset stands on the shoulders of their valuable work in Tibetan language preservation and documentation.

📜 License

This dataset is released under CC-BY-4.0 license. Please refer to the original source datasets for their specific licensing terms.

🤝 Contributing

Found an issue or want to contribute improvements? Please open an issue or pull request on the dataset repository.

📧 Contact

For questions, issues, or collaborations, please reach out through the Hugging Face discussion forum on this dataset page.

---

Last Updated: February 2025
Maintainer: lilgoose777
Version: 1.0